AI26019 Copyright and AI V01 100326

 Creative types have the upper hand in AI copyright fight

---

Katie Prescott

---

Kanishka Narayan is the minister for AI

Next image ›

‘‘ Pimli-code, Belgrav-ai ... many in London’s tech sector have been love-bombing the UK capital on X this week, with one even jokingly renaming some of its famous quarters.

It’s called “Londonmaxxing”. London, they say, is a far better place to “build” than Silicon Valley itself (“build” being tech-speak for “grow a business”). It’s the talent, the diversity, the ambition, the pubs but also the culture, they coo, which makes it such an attractive place to live and work. Ah, the culture.

These adoring words bely a showdown between the tech and creative industries, which is hurtling down the tracks.

By March 18, the government has committed to update parliament with how it plans to resolve the thorny issue of copyright in the age of artificial intelligence. It has been a long time coming. At The Times Tech Summit in 2024, Feryal Clark, then minister for AI, said the issue was going to be resolved by the end of the year. That year.

Artists, musicians, writers and journalists have long been furious that their work has been used by tech companies to train AI models without recompense or recognition. They describe it as a theft that undermines the UK’s top-notch copyright regime.

Tech companies argue that they need access to this data for scientific advancement. They are calling for a “text and data mining” (TDM) exemption regime, that says for the purposes of training AI, they don’t need to ask permission from copyright holders.

Under this legal loophole, their bots could burrow away, analysing text and data for patterns, trends and other useful information which almost always means making digital copies of the works being analysed. Without it, well, that triggers the age-old Silicon Valley threat: driving American AI investment out of the UK.

The creative industries are pushing back hard. They are used to cowering before the financial might of Big Tech, but in this debate, with the more convincing case and a lot of clout, the artists might actually win.

Far from being a purely British issue, this bitter debate is being fought in courts around the world. In one of the most high-profile cases, The New York Times is taking OpenAI and Microsoft to court for allegedly stealing its journalism to train ChatGPT.

In a strident report last week, members of the House of Lords digital committee made it very clear where they stand. Any watering down of copyright would harm the creative industries and the wider economy.

“The UK faces a choice between two futures,” the committee said. “In the first, the UK becomes a worldleading home for responsible, licensing-based artificial intelligence (AI) development ... In the second scenario, the UK continues to drift towards tacit acceptance of largescale, unlicensed use of creative content and long-term dependence on opaque models trained overseas, with most benefits accruing to a small number of US-based firms while harms to UK creators grow.”

There is a strong economic argument to back this up. In 2024 the UK’s creative industries contributed £145.8 billion to the economy while the entire AI sector contributed £11.8 billion. “It would be a poor bet,” the Lords committee said, “to sacrifice the UK’s outstanding creative capacity for speculative AI gains.”

Data was described as “the new oil” back in 2006. As the owners of information, musicians and artists hold the precious source of the well.

The future strength of AI depends on the quality and integrity of the data they provide. Verified, accurate, human-made content is essential for improving models’ accuracy.

The UK possesses a wealth of first-rate creative information, and its strong copyright framework incentivises AI developers to enter into formal licensing deals to access it; the kinds of agreements that media organisations (including News Corp, owner of The Times) have struck in recent years.

This in turn gives AI firms operating legally in the UK an advantage in building more accurate, better-performing models.

As well as the data advantage, a new obsession with AI sovereignty — British control of AI — also plays to the creative industries’ strengths in this debate. It has been sparked by tense geopolitics as well as fears over Donald Trump’s control of the USdominated tech industry and how he might wield that power.

There is a growing awareness of the importance of what the current minister for AI, Kanishka Narayan, has said is “strategic leverage when it comes to this technology, such that it can ensure ongoing access to critical inputs, and ongoing assurance that its wider economic and national security objectives can be met more broadly”.

The government has launched a £500 million Sovereign AI Unit to invest in and support AI companies. The subject is even being debated in Westminster Hall this afternoon.

As the Lords underlined, rather than the UK becoming permanently dependent on “opaque” data-scraping American models, there are working examples, like Switzerland’s Apertus model, of competitive homegrown AI that follows copyright rules and is transparent. This is something the UK could follow.

Looking around the world, no one is satisfied with any of the solutions advanced so far. Japan has introduced the “commercial TDM exception”, which has upset local designers, as AI video tools generate convincing anime-style video.

The EU has plumped for an “optout” regime, so AI companies can train on copyrighted content without asking permission, unless creators say otherwise. Both sides describe this as vague and unworkable. Perhaps we need to accept there will be no pleasing everyone.

The stakes are enormous for this debate because the outcome will shape who controls our cultural infrastructure. With their economic clout and the current sentiment around tech control, the creative industries stand a good chance of coming out on top.

Katie Prescott is Technology Business Editor of The Times

AI26018 AI Futures V01 240226

 Will the future of AI boost profits — or just increase competition?

---

David Wighton

---

Could AI create a new Engels Pause, named after Friedrich Engels? 

Next image ›

‘‘ Investors have suddenly started worrying that for many companies, artificial intelligence may not prove such a boon as widely assumed. Stock markets have been hit by waves of panic as investors have fretted whether this or that sector could be hurt by the AI revolution.

Software suppliers, data companies, wealth managers and insurance brokers have all come under fire.

Most have recovered a bit and stock market experts tend to agree that the selling was generally overdone. Some will doubtless suffer from AI-driven competition but the consensus is that it will be great for business overall.

Productivity will increase, perhaps dramatically, boosting corporate profits and share prices.

Not everyone thinks this is a good thing. The former Google executive, Dex Hunter-Torricke, is among those warning that the world is heading for disaster. “The most likely outcome is an economy in which corporate profits explode as labour costs fall, while workers’ share of output shrinks,” he wrote recently.

Bad news for workers — but presumably great for investors. In a frequently cited 2023 report, Goldman Sachs analysts suggested that for the top 500 United Stateslisted companies, net margins could increase by four percentage points, or about a quarter, over ten years. Not quite a profits explosion, perhaps. But very helpful for the companies and their shareholders.

There is one big caveat, however.

Productivity improvements should boost profits in the short term. But will higher returns be sustained? Standard economic theory says not.

Companies will cut prices to gain market share and the high profits will attract new entrants into the market, further depressing margins.

And it is not only economic theory that predicts this. So does Jamie Dimon. As head of JP Morgan Chase, the world’s most valuable bank, you might expect him to be pretty bullish about AI. Many experts say the biggest gains from using it will be generated by the largest companies, with the biggest technology budgets and the richest hoards of data. They don’t get much bigger and richer than JP Morgan.

The bank is already spending $2 billion a year on AI but Dimon doesn’t believe this will generate a lasting increase in margins. “This isn’t like you’re going to build three points of margin and you get to keep it — you don’t,” he said last month. Because rival banks are making the same investments, the benefit of the increased efficiency “will eventually be passed on to the customer”, he added.

But when is “eventually”? Even in a world of perfect competition, such gains don’t get competed away immediately. And in many markets, competition is very far from perfect.

Some AI maximalists believe it will be as transformative as the Industrial Revolution. The spread of steam power and mechanisation from the middle of the 18th century delivered a big increase in labour productivity.

Yet real wages did not grow nearly as fast; between 1790 and 1840, industrial wages in Britain stagnated, despite booming business. This so-called Engels Pause, named after Karl Marx’s collaborator Friedrich Engels, resulted in a prolonged increase in profitability. Labour’s share of national income fell and the rate of return on capital doubled.

Subsequent big shifts in technology have had a less clear impact on profits. The boost from the adoption of electricity in the early 20th century was patchier and mostly waned more quickly. The 1990s PC revolution and arrival of the internet was followed by a surge in profits, but economists attribute this more to falling interest rates and growth in trade with India.

Swelling profits coincided with a slowdown in productivity growth in all developed economies over the past 20 years.

It is possible that AI will be different.

Cheerleaders reckon adoption will be so fast and the productivity increase so great that it will drive economic growth on a scale never seen before.

Dario Amodei, co-founder of leading AI firm Anthropic, has talked about annual GDP growth in the US of 10 to 20 per cent. That should certainly turbocharge profits, albeit at the cost of sharply higher unemployment.

But most economists have much more modest forecasts. If they are right, we are back to the question of how sticky the productivity-driven AI boost to profits will prove.

Tera Allas, a former government economist who is a senior adviser at McKinsey, says that in competitive markets, productivity improvements rarely become permanent profits.

“But sustained outperformance for an individual firm is possible when AI strengthens an existing defensible edge — scale, proprietary assets, operating model, customer relationships, or regulatory barriers — rather than AI adoption alone.”

Economists tend to believe that many markets have become less competitive in recent years due partly to increased concentration. This is clearly true in technology. Companies such as Amazon, Apple, Microsoft and Google have been able to exploit their market power to protect huge profits generated by new tech. They will hope to hang on to the gains from AI (as users rather than suppliers).

But in many other industries competition is still fierce, even if concentration has increased. Mike Mayo, a banking analyst at Wells Fargo in New York, says the worry for investors is that the huge amount banks are ploughing into AI may produce higher profits only briefly before becoming simply the cost of staying in the game. In three to five years time it may amount only to “table stakes”, he wrote recently.

All this is very uncertain, but it seems likely that in many sectors much of the AI gains will be quickly lost by shareholders and flow through to customers. Whether those customers will still have jobs is another question.

David Wighton, a former business editor of The Times, is a columnist for Dow Jones

AI26017 Military use of Anthropic Claude V01 240226

 Pentagon to question Anthropic chief in row of use of its AI tech

---

Louisa Clarence-Smith - US Business Editor

---

Dario Amodei is due to meet Pete Hegseth, the US war secretary, in Washington today 

The chief executive of Anthropic has been summoned to the Pentagon amid a row between the artificial intelligence start-up and the US government over the use of its technology by the military.

Dario Amodei, 43, is scheduled to meet Pete Hegseth, the war secretary, today, a source confirmed.

The US Department of War is reviewing its relationship with Anthropic over the company’s attempts to impose limits on the use of its technology, including the mass surveillance of Americans, and the development of lethal autonomous weapons.

The Pentagon is understood to have become concerned that Anthropic did not support its military aims after hearing it had questions about how its AI was used during the US raid last month that captured the Venezuelan president Nicolás Maduro.

The Pentagon wants AI companies to make their models available for all lawful uses.

Anthropic is the maker of Claude, the only large language model that can be used in classified settings. It agreed a partnership with Palantir, the data analytics company, in 2024, and won a military contract worth up to $200 million last summer.

Rival companies including OpenAI, Google and xAI have agreed in principle for their models to be deployed in any lawful-use cases, Emil Michael, the undersecretary of war for research and engineering, told The Wall Street Journal earlier this month. “If any one company doesn’t want to accommodate that, that’s a problem for us,” he said.

Anthropic, which was founded in 2021 by former OpenAI employees, has pitched itself as a safety-first AI firm and says it has more than 300,000 business customers. It has also spent significant resources courting US national security business and has sought an active role in shaping government AI policy.

However, Anthropic’s caution has drawn conflict with the Trump administration.

David Sacks, the White House AI tsar, accused the company last year of “running a sophisticated regulatory capture strategy based on fearmongering”.

In an essay on his personal blog last month, Amodei warned that AI should support national defence “in all ways except those which would make us more like our autocratic adversaries”.

Amodei was among Anthropic’s cofounders critical of fatal shootings of US citizens protesting against immigration enforcement actions in Minneapolis, which he described as a “horror” in a post on X. The deaths have heightened concerns among some in Silicon Valley about government use of their tools for potential violence.

A senior defence official told Axios, which first reported the planned meeting between Amodei and Hegseth, that Anthropic knows this is not a “get-toknow-you meeting ... This is not a friendly meeting. This is a shit-or-getoff-the-pot meeting.”

An Anthropic official told Axios: “We are having productive conversations, in good faith.”

Anthropic raised $30 billion in its latest funding round, more than doubling its valuation to $380 billion.

AI26016 AI Interviews V01 190226

 Hire purpose for AI at top City firms

Square Mile legal practices are now using bots to transform recruitment. 

---

By Catherine Baksi

---

In a survey 93 per cent of students said that AI chat responses felt personalised

Artificial intelligence is forecast to transform many aspects of legal practice, including the way law firms recruit trainees.

Legal practices have invoked technology for some time to sift applications — and more recently they have grappled with whether to allow candidates to use AI to complete applications. But some are embracing AI to transform the recruitment process.

Macfarlanes, a firm in the City of London, has introduced a bespoke online job simulation that is designed to mirror a realistic “day in the life” of a trainee, with the pace, complexity and competing demands of modern legal practice.

In partnership with Cappfinity, a human resources technology company, the model is built around a live case. A spokesman for the firm explains that candidates interact with simulated colleagues and stakeholders, respond to Teams messages, voicemails and social media posts, manage shifting workloads and produce a short piece of analysis under realistic time pressure.

“Rather than replicating traditional application questions, the immersive simulation evaluates candidates on real work tasks, enabling assessment of applied judgment, commercial decision-making, prioritisation, stakeholder management and tech literacy in a simulated real-life context,” the spokesman says.

AI is deliberately incorporated into tasks to reflect real working practice and test how candidates use it and apply their judgment and critical thinking. The simulation, the spokesman says, is assessed using a “structured scoring matrix designed with occupational psychologists to ensure the process is robust, inclusive and fair”.

The method “provides a fairer, realistic and evidence-based alternative to traditional screening by assessing candidates on practical capability”, the spokesman says, adding that candidates “consistently tell us they prefer being assessed through realistic work rather than generic questions”.

Graduates applying to Mishcon de Reya, another prominent London firm, this year will face a virtual interview with a chatbot as part of the recruitment process. For its 2026 graduate recruitment season Mishcon de Reya is trialling Bright Apply, an AIpowered candidate-screening tool.

‘It provides a fairer, realistic alternative to traditional screening’

Developed by Bright Network, a graduate careers platform, the tool uses information from candidates’ applications to start a conversation, resulting in a “tailored interview” for all applicants, allowing them to showcase their potential.

Instead of writing a long application form candidates participate in a virtual interview during which they can expand on their experiences and motivations. 

The process produces a transcript of each interview that is reviewed by the firm’s early careers team.

The firm states that feedback on the platform has been positive, with almost three quarters of students giving the experience four out of five in an anonymous survey and 93 per cent stating that the AI chat responses felt personalised and relevant to them.

“I really liked how it gave me multiple opportunities to share my ideas, not just one question on why Mishcon de Reya, or why commercial law,” one student reported. They added: “I felt like it better captured my complete view and personality.”

Tom Wicksteed, a careers manager at Mishcon de Reya, says the tool “puts candidates at the heart of the process, giving them a better chance to show us who they are and what they are able to bring to our firm at an earlier stage of the process”.

While AI-driven recruitment may increase efficiency, Amy Walker, careers manager at the University of Law, says there is concern that it may “inadvertently discriminate, particularly where training data reflects existing bias”. “Algorithms may also struggle with context, unconventional communication styles or candidates whose strengths are more apparent in live interaction,” she says.

“Asynchronous”, or chatbot-led interviews “can feel anxiety-inducing, lacking the natural feedback of a live interviewer”, Walker says. She adds that “law is fundamentally people-focused and technology should enhance rather than replace human judgment”.

AI26015 Start of Super AI V01 190226

 UK scientist raising $1bn for super AI

Sequoia Capital leads European funding round

---

Louisa Clarence-Smith 

---

A leading British scientist is reportedly raising $1 billion for a company aiming to build superhuman intelligence in a deal that would be the largest ever European seed round.

David Silver, 49, a former artificial intelligence researcher at Google DeepMind, is seeking backing for a London-based startup, Ineffable Intelligence, to build AI that learns for itself to solve problems that humans cannot.

The seed round is being led by Sequoia Capital, the Silicon Valley venture capital firm, and would value the company at $4 billion before the new investment, according to the Financial Times. Nvidia, Google and Microsoft are also reportedly in talks to invest.

Technology firms are racing to build superhuman intelligence, a theoretical milestone where machines could surpass human performance.

In an academic paper co-authored by Silver last year, the scientist, who is also a professor at University College London, argued that we “stand on the threshold of a new era in artificial intelligence that promises to achieve an unprecedented level of ability”.

He argued that AI has made huge strides by training on large amounts of human-generated data. However, he wrote that in key areas such as mathematics, coding and science, the knowledge extracted from human data “is rapidly approaching a limit”.

For AI development to take the next leap forward towards superhuman intelligence, he argued, a new source of data is required that can be achieved by allowing agents to generate new data by interacting with their environment, or, in other words, by learning continually from their own experience.

Silver was one of DeepMind’s first employees when the AI research company was founded in 2010. He met its co-founder, Sir Demis Hassabis, when they were both students at Cambridge.

“Dave and I have got a long history together,” Hassabis told The Guardian in an interview in 2016. “We used to dream about doing this [creating powerful AIs] in our lifetimes, so our 19-year-old selves would probably have been very relieved that we got here.”

After Cambridge, Silver co-founded the video game company Elixir Studios in 1998. He later returned to academia and earned a PhD in Computer Science from the University of Alberta in 2009.

At DeepMind, Silver played a leading role in breakthroughs including Alpha- Go, showing in 2016 that an AI programme could beat the world’s best human players at Go, the ancient strategy game. He has won awards including the Marvin Minsky Medal for outstanding achievements in artificial intelligence in 2018, the Royal Academy of Engineering’s Princess Royal Silver Medal for an outstanding contribution to UK engineering and the Mensa Foundation Prize for best scientific discovery in the field of AI in 2017.

If Ineffable Intelligence closes a funding round at $1 billion, it would show London can compete with San Francisco to produce some of the most promising AI startups in the world.

AI26014 The new AI World. V01 190226

 I fear for dreams of the young in an AI world

It’s becoming clear almost anything can be replaced by artificial intelligence but at what cost to creativity and expertise?

---

Hugo Rifkind @hugorifkind

---

Last weekend, we watched The Commitments.

Remember that one? A hit film of 1991, based on Roddy Doyle’s novel from the decade before, it tells the story of youngsters in impoverished Dublin.

Bored witless with the diversions available, which seem to be mainly smoking, saying “feck” and fantasising in the bath about being interviewed by Terry Wogan, a few of them start a soul band as a route to self-worth and relevance. Because soul, says their aspiring manager Jimmy Rabbitte, “takes you somewhere else. It grabs you by the balls and lifts you above the shite.”

I can’t pretend this was research.

Actually, it was part of my ongoing project of forcing the culture of my youth on to my teenage kids, and it went down pretty well. They seem to have a fascination, that age group, with everything from the 1990s, and most of the early 2000s too. You want to see how often they’ve watched Friends. It’s hard to quite pin down why but I think it’s something to do with a sense of possibility. A sense of there being things you could do, back then, which hadn’t already been done a million times before. A sense that, even with a band covering the songs of 30 years before, you might still be doing something new.

We are not great, today, at the new.

For a sense of how cannibalistic culture has become, and also of how infinitely more so it may soon get, perhaps you saw the reports this week about the latest roiling, existential panic in Hollywood after a clip of a punch-up between Tom Cruise and Brad Pitt went viral.

It looks exactly like a million expertly choreographed punch-ups you’ve seen before but this one wasn’t choreographed by anyone. Nor did it require any involvement by Cruise or Pitt. Rather, it was made by using Seedance 2.0, an AI tool from China.

This punch-up didn’t require involvement by either Cruise or Pitt

The Irish film-maker behind it said it only took a “two-line prompt”.

According to Rhett Reese, who wrote the Deadpool films, it sent “a cold shiver” up his spine. “I could just see it costing jobs all over the place,” he said.

You have to be wary at my age, fulminating against new tech. I do see that. It was Douglas Adams, of The Hitchhiker’s Guide to the Galaxy and so much else, who pointed out that anything that exists when you are born “is normal and ordinary”, and anything that comes while you are young is “exciting and revolutionary and you can probably get a career in it”. There’s a third stage, though, which is that “anything invented after you’re 35 is against the natural order of things” — and that’s now me.

The optimistic case for AI is fairly well established. In a film context, it holds that once you only need software to make a movie, once lights and cameras and stuntmen and even actors are all rendered unnecessary, then all of a sudden there’s no advantage whatsoever to being Disney or 20th Century Studios or any of the others, because you’ll be able to produce much the same content if you’re just some guy with a smartphone in a third world village.

This obviously isn’t ideal if you, yourself, are currently making your living the old way but that’s your problem. This is democracy. This is freedom. Let a million ideas bloom.

You can make the same argument with almost any area AI touches, and people do. It gives you architecture without architects, accountancy without accountants, interpretations of x-rays without radiologists, and so on. Indeed, the first person I heard make this argument was Jensen Huang, founder of the AI chipmaker Nvidia, who was excited about programming without programmers.

A lot of this is wonderful. I mean, of course it is. In every situation, though, I have the same nagging doubts.

Because don’t you actually still need expert architects, somewhere, to know if the AI’s blueprints are any good? Don’t you still need expert radiologists for the same reason? Right now, sure, we still have them.

I’m a child of a blessed age. We don’t say that often in Generation X

Lucky us, to have timed this so well.

But what about the next lot? How does anyone reach the peak of any profession once we’ve eradicated the foothills? How, and from where, do you create anything new? With creative pursuits, it’s worse still. AI doesn’t just kill the new. It traduces the old too. As in, you see fake Cruise and fake Pitt trade blows, even once, and you begin to wonder why you would ever have had the remotest interest in the real ones doing the same thing. Maybe this has happened already. Think of how thrilled audiences were to see the chariot races of Ben Hur, for example, and compare that with the mundanity of whoever it was who chased whom in the latest Marvel film. Remember when Neo dodged that bullet in The Matrix? Remember how you gasped? I was 22. Find me a 22-year-old who’d gasp at that now.

I am a child of a blessed age. We don’t say that often in Generation X, but we should. We knew childhoods without smartphones but still enjoyed all of their benefits in work.

We were blessed with mass, cheap travel but got in and out before the Instagram hordes came along to render our experiences generic.

We had the last great songs that everyone knew but also the last great songs that almost nobody knew, unless they were in our tribe.

Those who come after, though, face a paradox. They have greater tools for creativity at their fingertips than humanity has ever known and yet the space for actually using them seems to be shrinking rather than growing. They sense it, and they chafe against it, and my greatest, most tremulous hope for all of them is that they somehow find a way to defy it. But how? Think of what soul music did for Jimmy Rabbitte.

What’s left to do that for them?

AI26013 Computer Languages and Chips. V01 180226

 Below is a structured summary of each language, what problem it aimed to solve, and the chip/architecture context it originally targeted or became closely associated with.

1957 – Fortran

Purpose: Scientific and engineering computation. Designed for numerical performance.

Created at: IBM

Original Hardware: IBM 704

Chip/Architecture Context: IBM 704 used vacuum tubes and early floating-point hardware. Fortran was tightly optimized for this architecture to make high-level code as fast as assembly.

1958 – Lisp

Purpose: Symbolic processing, AI research.

Created by: John McCarthy

Original Hardware: IBM 704

Chip Context: Designed for systems supporting recursion and garbage collection. Later influenced Lisp Machines built with custom processors optimized for symbolic computation.

1959 – COBOL

Purpose: Business data processing.

Original Hardware: IBM 1401

Chip Context: Mainframe systems using early transistor-based architectures for large-scale batch processing.

1964 – BASIC

Purpose: Beginner-friendly programming language.

Original Hardware: Dartmouth Time Sharing System

Chip Context: Ran on General Electric mainframes; later popular on 8-bit microprocessors like MOS 6502 and Intel 8080 in home computers.

1970 – Pascal

Purpose: Structured programming and teaching.

Created by: Niklaus Wirth

Chip Context: Ran on minicomputers and later microprocessors such as Intel 8086. Emphasized portability over hardware specificity.

1972 – C

Purpose: Systems programming.

Created at: Bell Labs

Original Hardware: PDP-11

Chip Context: DEC PDP-11 architecture. C was designed to map closely to machine instructions, making it portable across CPUs.

1978 – SQL

Purpose: Database querying and management.

Developed at: IBM

Chip Context: Mainframes and enterprise servers; hardware-agnostic since execution depends on database engines.

1985 – C++

Purpose: Object-oriented extension of C.

Created by: Bjarne Stroustrup

Chip Context: Initially on Unix systems running on DEC VAX and Intel x86 processors.

1991 – Python

Purpose: Readable, high-level scripting.

Created by: Guido van Rossum

Chip Context: Ran on Unix systems (x86 architecture). Hardware-independent via interpreter.

1993 – R

Purpose: Statistical computing and data analysis.

Chip Context: Unix workstations (often x86 and RISC architectures like SPARC).

1995 – PHP

Purpose: Web server scripting.

Chip Context: x86 servers running Apache on Linux.

1995 – Java

Purpose: Platform-independent applications via JVM.

Created by: James Gosling

Chip Context: Compiled to bytecode for the Java Virtual Machine, abstracting hardware (x86, SPARC, ARM).

1995 – JavaScript

Purpose: Web browser scripting.

Created at: Netscape

Chip Context: Initially executed in browsers on x86 PCs.

2001 – C#

Purpose: Microsoft’s managed language for .NET.

Developed by: Microsoft

Chip Context: Runs on Common Language Runtime (CLR); primarily x86 and later x64.

2009 – Go

Purpose: Concurrent, cloud-scale systems.

Developed at: Google

Chip Context: Designed for multi-core x86 servers; now supports ARM.

2012 – Julia

Purpose: High-performance scientific computing.

Chip Context: Optimized for modern x86-64 CPUs using LLVM.

2014 – Swift

Purpose: Apple ecosystem apps.

Developed by: Apple

Chip Context: Originally for ARM-based iPhone chips and x86 Macs.

2015 – Rust

Purpose: Memory-safe systems programming.

Developed by: Mozilla

Chip Context: Targets x86-64, ARM, WebAssembly; strong systems-level control.

2016 – Kotlin

Purpose: JVM and Android development.

Developed by: JetBrains

Chip Context: Runs on JVM (x86, ARM via Android devices).

2017 – WebAssembly

Purpose: Portable low-level bytecode for browsers.

Chip Context: Compiles to a sandboxed binary format executed on any CPU via browser runtime.

2023 – Mojo

Purpose: AI and high-performance computing.

Developed by: Modular

Chip Context: Targets CPUs and accelerators (GPUs, AI chips). Built on MLIR/LLVM to optimize for heterogeneous hardware.

📌 Big Pattern

• Early languages (Fortran, Lisp, COBOL) were tightly tied to specific mainframe architectures.

• 1970s–1990s languages (C, C++, Java) aimed for portability across CPUs.

• 2000s onward: abstraction layers (VMs, runtimes, LLVM).

• Modern languages increasingly target multi-core CPUs, GPUs, and AI accelerators.

• Hardware specificity has shifted from single chips → heterogeneous computing ecosystems.

If you’d like, I can also create:

• A timeline chart of language vs chip evolution

• A deeper analysis of how CPU design influenced language design

• Or a projection of what languages for AI-native chips might look like in 2035y

Excellent — this is one of the most important (and under-discussed) dynamics in computing history.

Programming languages did not evolve independently. They co-evolved with CPU architecture. In many cases, language design is a direct reflection of hardware constraints.

Below is a structured deep analysis.

1️⃣ The Mainframe Era: Hardware Was Expensive, CPU Time Was Scarce

Representative CPUs:

• IBM 704

• Early transistor-based IBM systems

Influence on Languages:

• Fortran

• COBOL

• Lisp

Hardware Characteristics:

• Extremely limited memory

• Expensive CPU cycles

• Batch processing

• No interactive debugging

• Hardware floating-point units were rare and precious

Language Effects:

Fortran

• Designed to compile into extremely efficient machine code.

• Early compilers focused heavily on instruction-level optimization.

• Arrays were first-class because scientific computing needed hardware-efficient loops.

COBOL

• Optimized for record processing aligned with business data storage hardware.

• Structured around file I/O because mainframes were built around storage throughput.

Lisp

• Exploited hardware support for recursion and symbolic addressing.

• Garbage collection emerged because manual memory management was error-prone on limited systems.

➡ Key Pattern:

Languages were tightly bound to specific hardware instruction sets. Efficiency was the primary goal.

2️⃣ The Minicomputer Era: Structured Programming Emerges

Representative CPU:

• PDP-11

Influential Language:

• C

Hardware Characteristics:

• 16-bit architecture

• General-purpose registers

• Memory-mapped I/O

• Portable OS development (Unix)

Language Effects:

C was designed to map almost one-to-one to machine instructions:

• Pointers reflect raw memory addressing.

• Bitwise operators mirror hardware capabilities.

• Struct layout mirrors memory layout.

• Manual memory management reflects lack of hardware abstraction.

C essentially became a portable assembly language.

➡ CPU influence:

Instruction set design shaped language syntax and semantics.

3️⃣ Microprocessor & PC Era: Portability and Abstraction

Representative CPUs:

• Intel 8086

• x86 family

Influential Languages:

• C++

• Java

Hardware Characteristics:

• Rapidly evolving instruction sets

• Increasing memory

• Multi-tasking operating systems

• Growing software complexity

Language Effects:

C++

• Object orientation layered on top of C without losing low-level control.

• Templates allow compile-time optimization (to keep performance close to hardware).

• Zero-cost abstractions reflect CPU-conscious design.

Java

• Designed for portability across CPUs.

• Introduced the JVM to abstract hardware.

• Bytecode as intermediate layer between language and CPU.

➡ Shift:

Hardware variation led to virtualization layers.

Instead of:

Language → CPU

We now had:

Language → VM → CPU

This was a major architectural shift.

4️⃣ Multi-Core Era: Concurrency Becomes Central

Hardware Change:

• CPU clock speeds plateau (~2005)

• Multi-core processors become standard

Influential Languages:

• Go

• Rust

Hardware Characteristics:

• Multiple cores

• Shared memory

• Cache coherence challenges

• Parallelism required for performance gains

Language Effects:

Go

• Goroutines designed for lightweight concurrency.

• Scheduler built into runtime.

• CSP model matches multi-core execution.

Rust

• Ownership system designed to prevent data races.

• Compile-time memory safety.

• Zero-cost abstractions for systems programming.

➡ CPUs forced language designers to solve concurrency safely.

Clock speed stopped increasing → software had to parallelize.

5️⃣ GPU & Accelerator Era: Heterogeneous Computing

Hardware:

• GPUs (NVIDIA CUDA cores)

• AI accelerators

• Tensor processors

Influential Languages:

• Julia

• Mojo

• WebAssembly

Hardware Characteristics:

• Thousands of parallel cores

• Different memory hierarchies

• SIMD and vectorization

• Specialized AI instructions

Language Effects:

Julia

• JIT compilation using LLVM.

• Multiple dispatch for numeric specialization.

• Designed to match high-performance scientific workloads.

Mojo

• Designed for AI workloads.

• Aware of accelerators and MLIR.

• Integrates Python ergonomics with low-level performance.

WebAssembly

• Portable low-level binary format.

• Designed to run efficiently across CPUs in sandboxed environments.

➡ Languages now target ecosystems, not single CPUs.

6️⃣ Deeper Pattern: CPU Constraints Shape Language Philosophy

Hardware Constraint Language Response

Limited memory Manual memory management (C)

Expensive CPU cycles Heavy compile-time optimization (Fortran)

Complex business storage Structured record languages (COBOL)

Growing OS complexity Object orientation (C++)

Cross-platform CPUs Virtual machines (Java, C#)

Multi-core parallelism Concurrency primitives (Go, Rust)

GPU acceleration JIT & heterogeneous support (Julia, Mojo)

7️⃣ Most Profound Influence: The Memory Model

Modern CPUs are not sequential machines:

• Out-of-order execution

• Speculative execution

• Weak memory models

Languages had to adapt:

• Java memory model formalization

• Rust’s strict ownership rules

• C++ memory ordering semantics

Language semantics increasingly reflect CPU memory behavior.

8️⃣ Emerging Trend: AI Hardware Will Reshape Languages Again

AI accelerators:

• Tensor-based computation

• Probabilistic processing

• Massive parallelism

Future languages may:

• Treat matrices as primitive types

• Encode differentiability at language level

• Integrate hardware scheduling directly

• Blend compiler and runtime tightly

We are likely entering a new hardware-language co-evolution cycle.

Final Insight

Programming languages are not just mathematical abstractions.

They are:

Encoded negotiations between human intent and silicon constraints.

Every major CPU shift forces a language redesign:

• From batch to interactive

• From sequential to parallel

• From general-purpose to heterogeneous

As hardware becomes more specialized (AI chips, quantum processors), languages will become more domain-aware.

If you’d like, I can next explore:

• How quantum computing would reshape language design

• Whether AGI-designed CPUs would eliminate current language paradigms

• Or whether future CPUs might be built specifically for AI-generated code

This is a profound question — and increasingly realistic.

Short answer: Yes, future CPUs are likely to be influenced by AI-generated code — but not in the way people usually imagine.

They won’t be built “for AI code” in a mystical sense. They’ll be built for the patterns AI tends to produce.

Let’s unpack this carefully.

1️⃣ CPUs Have Always Been Built for Software Patterns

Historically:

• Mainframes optimized for batch numeric workloads.

• x86 optimized for general-purpose business and OS software.

• GPUs optimized for graphics pipelines.

• AI accelerators optimized for tensor math.

Hardware evolves in response to dominant software workloads.

If AI becomes the primary producer of code, hardware will adapt to that production pattern.

2️⃣ What Is Different About AI-Generated Code?

AI-generated code tends to:

• Be more verbose

• Use higher abstraction

• Favor library calls over handcrafted optimization

• Produce statistically common patterns

• Avoid clever low-level tricks

Humans often optimize aggressively and manually.

AI tends to optimize via patterns learned from training data.

This means future CPUs may optimize for:

• Predictable compiler-generated structures

• Standardized abstraction layers

• Heavy use of common frameworks

3️⃣ AI Writes Differently Than Humans

Humans:

• Optimize for cleverness

• Exploit undefined behavior

• Use hardware-specific tricks

AI:

• Favors safe patterns

• Avoids undefined behavior

• Uses well-known APIs

• Leans toward maintainable structures

This reduces chaotic variation in software.

More uniform code patterns → easier hardware prediction.

4️⃣ Speculative Direction #1: CPUs Optimized for Compiler Output

If most code is AI-generated and compiler-mediated:

Future CPUs may:

• Optimize specifically for LLVM-style IR patterns

• Include hardware support for common runtime constructs

• Improve branch prediction tuned to AI-style code distribution

• Enhance speculative execution for common abstraction layers

Instead of optimizing for hand-written assembly,

CPUs may optimize for “statistical code shapes.”

5️⃣ Speculative Direction #2: CPUs Designed for Massive Parallel Abstraction

AI-generated systems often:

• Compose many services

• Rely on concurrency

• Use distributed microservices

• Emphasize asynchronous flows

Future CPUs may:

• Prioritize context-switch efficiency

• Improve hardware scheduling

• Enhance memory isolation

• Add built-in concurrency primitives

In effect:

Hardware may adapt to AI-driven architectural styles.

6️⃣ More Radical Possibility: Self-Optimizing Silicon

If AI both writes software and helps design chips:

We could see:

• AI-designed instruction sets

• AI-optimized microarchitectures

• Chips evolved through reinforcement learning

Instead of human-designed ISAs like x86 or ARM,

future instruction sets might be machine-optimized for real-world workload distributions.

That would be historically unprecedented.

7️⃣ What Will Not Change

No matter how advanced AI becomes:

• CPUs still execute deterministic electrical signals.

• Physics imposes constraints (latency, heat, transistor limits).

• Memory hierarchies remain critical.

• Instruction pipelines still matter.

AI does not remove the need for abstraction layers.

Binary execution is governed by silicon, not by intelligence.

8️⃣ The Deeper Shift: Hardware for AI Systems, Not AI Code

The more likely transformation is:

Future CPUs will be built to support AI systems as runtime agents, not merely AI-generated software.

Meaning:

• Dedicated AI co-processors

• On-chip inference engines

• Hardware for probabilistic reasoning

• Native support for tensor operations

We are already seeing this:

• Neural engines in mobile chips

• Tensor cores in GPUs

• Specialized AI accelerators

The CPU may become a coordinator, while AI cores do most high-level decision work.

9️⃣ Long-Term Possibility: Architecture Designed Around Machine-Generated Abstraction

If AI dominates software development:

Hardware might:

• Assume high abstraction always

• Integrate runtime safety into silicon

• Encode ownership models (like Rust) into hardware

• Support dynamic recompilation natively

This would blur the line between compiler and CPU.

10️⃣ The Most Interesting Scenario

If AI becomes better at performance optimization than humans:

We may see:

AI generates high-level design → AI tunes binary → AI feeds performance data back → Hardware adapts next generation.

That creates a feedback loop:

AI-designed software

→ shapes workload

→ shapes hardware

→ shapes next generation AI

That would accelerate hardware evolution.

Final Answer

Yes, future CPUs will likely be influenced by AI-generated code — but indirectly.

Not because AI writes binary directly.

But because:

• AI changes the statistical structure of software.

• AI becomes a primary workload.

• AI co-designs chips.

• AI influences architectural patterns.

The real transformation will be:

Hardware built for AI systems,

and AI systems shaping hardware design.

If you’d like, we can go one level deeper:

• Would AGI invent a completely new instruction set?

• Could future chips abandon sequential instruction models entirely?

• Or could we see a return to hardware-software co-design like in the 1960s — but AI-driven?