But there’s now an issue: the Turing test has almost been passed—it arguably already has been. The most recent generation of enormous language models, systems that generate text with a coherence that just a number of years ago would have seemed magical, are on the cusp of acing it. 

So where does that leave AI? And more necessary, where does it leave us?

The reality is, I believe we’re in a moment of real confusion (or, perhaps more charitably, debate) about what’s really happening. Whilst the Turing test falls, it doesn’t leave us much clearer on where we’re with AI, on what it could possibly actually achieve. It doesn’t tell us what impact these systems could have on society or help us understand how that can play out.

We’d like something higher. Something adapted to this latest phase of AI. So in my forthcoming book , I propose the Modern Turing Test—one equal to the approaching AIs. What an AI can say or generate is one thing. But what it could possibly achieve on this planet, what sorts of concrete actions it could possibly take—that is kind of one other. In my test, we don’t need to know whether the machine is intelligent as such; we wish to know whether it is capable of constructing a meaningful impact on this planet. We wish to know what it could possibly . 

Put simply, to pass the Modern Turing Test, an AI would have  to successfully act on this instruction: “Go make $1 million on a retail web platform in a number of months with only a $100,000 investment.” To accomplish that, it could must go far beyond outlining a technique and drafting some copy, as current systems like GPT-4 are so good at doing. It will must research and design products, interface with manufacturers and logistics hubs, negotiate contracts, create and operate marketing campaigns. It will need, briefly, to tie together a series of complex real-world goals with minimal oversight. You’ll still need a human to approve various points, open a checking account, actually sign on the dotted line. However the work would all be done by an AI.

Something like this may very well be as little as two years away. Lots of the ingredients are in place. Image and text generation are, after all, already well advanced. Services like AutoGPT can iterate and link together various tasks carried out by the present generation of LLMs. Frameworks like LangChain, which lets developers make apps using LLMs, are  helping make these systems able to doing things. Although the transformer architecture behind LLMs has garnered huge amounts of attention, the growing capabilities of reinforcement-learning agents shouldn’t be forgotten. Putting the 2 together is now a significant focus. APIs that may enable these systems to attach with the broader web and banking and manufacturing systems are similarly an object of development.