Another day, another claim by a frontier AI company to have achieved a huge breakthrough in science. Most recently, it’s Anthropic, saying:
‘Claude has discovered a previously unknown enzyme system hidden in the DNA of bacteriophages. Beside the enzyme’s gene sits a long array of repeating DNA – a structure that looks somewhat similar to CRISPR.
We don’t yet understand what this system does, but only a handful of known systems share its features, and all of them are able to cut, copy, and paste DNA. Historically, the discovery of such programmable systems has helped revolutionize medicine. CRISPR, for instance, is now the foundation of genetic medicines. But it will take much more work to learn what this system does, and whether it can be put to similar use.
In a post which has gained over 21 million views. Anthropic’s CEO Dario Amodei said of the discovery:
‘Its precise function, biotechnological utility (if any), or level of significance is not yet clear, but at minimum it is work I would have been proud to do as a PhD student.’
He went on to say this was proof that AI might be able to cure most diseases within 5-10 years, a prediction that he also made in October of 2024, so roughly within the next eight years.
‘This suggests that by 2030 people using AI services will have to be spending around $6 trillion a year by 2030, around 15 per cent of US GDP.’
Hidden enzymes and ‘CRISPR’ – it all sounds very promising. The discovery of CRISPR was an enormous leap forward in biotechnology as it enabled us to cheaply edit genes for the first time. The possible applications of CRISPR are wondrous. We could, as an example, eliminate malarial mosquito populations, as Target Malaria aims to do – though our current ethics and fears over genetic modification have prevented these more far-reaching applications from being explored in full, an important point when discussing speculative biotechnology in general.
But how significant is Anthropic’s discovery really? If you look beneath the framing you see that Anthropic let loose Claude agents on a huge database which had been created by lab work by human beings. It detected a pattern that nobody had noticed before, which merits some further investigation – but there has been software available for analysing these large datasets for years, predating Claude. Tools which can spot repeat arrays (what Claude detected), including PILER-CR, go back to the 2000s. As one researcher with a background in bioinformatics background put it:
‘That is a good result. But in a regular biology lab, this isn’t the finished paper. It is the result you show at lab meeting and say: “This looks interesting. Now we need to figure out what the hell it does.” Maybe that next step leads to a major discovery. But that discovery hasn’t happened yet.’
It is understandable that Anthropic is talking up the potential utility of its services ahead of an IPO in the coming months. That’s just good business. And who knows – maybe LLMs will dramatically transform medicine research. But Anthropic has not been able to prove that they can. But because the company has an effective PR machine, and politicians treat them as if they are magicians instead of salesmen, their claims are rarely challenged.
Which is a problem. The narrative that AI will solve almost all disease raises the stakes of the technology so high that other problems with it (like the fact that LLMs use content researchers and writers create without paying for it) seem mundane. This speculation over its potential power is also creating an enormous bubble. As Bloomberg points out, it is estimated that investment in AI infrastructure will rise to $1 trillion a year before 2030. This suggests that by 2030 people using AI services will have to be spending around $6 trillion a year by 2030, around 15 per cent of US GDP.
Plausible? Well if AI is so powerful that its users can find a cure to cancer with a two sentence prompt, perhaps. But if Amodei is wrong there will be a lot of money to lose, not least by the retail investors who buy into Anthropic at IPO stage.
Comments