One possible solution to this problem is the use of so-called confidence thresholds, i.e. the systems are programmed in such a way that they either only produce an answer if the systems are rather certain of it or they indicate how certain they are of the individual statements. In the case of deterministic AI – i.e. systems that specialise in recognising or classifying certain things – such values are commonly used. In the field of generative AI, however, this is not yet very common. In our view, it should be used more often. For example, a chatbot can be programmed so that it only provides an answer if it is relatively certain of it. It is, however, not clear how high the probability must be for something to be considered (supposed) fact instead of (supposed) fiction.

These insights don’t change the fundamental realities of the natural world — they reveal it to be something very different than what our intuitions and cultural cosmologies previously taught us. That revealing is the crux of the trauma. All the stages of grief are in response to the slow and then sudden fragmentation of previously foundational cultural beliefs. Like the death of a loved one, the death of a belief is profoundly painful.

The premise is that modern governments as we know them are the executives of the transformations to come and not an institutional form that will be overhauled if not absorbed by them. For better or worse, the latter scenario may be more plausible.

The leap of faith that human values are self-evident, methodologically discoverable and actionable, constructive, and universal is the fragile foundation of the alignment project. It balances on the idea that it will be possible to identify common concerns, to poll communities about their values and conduct studies about the ethics of possible consumer products, that it will be possible and desirable to ensure that the intelligence earthquake is as comfortable as possible for as many people as possible in as many ways as possible.

This stage of grief clings to the hope that if we start bargaining with the future then the future will have no choice but to meet us halfway. If only.

To what extent is the human artificialization of intelligence via language (as for an LLM) a new technique for making machine intelligence, and to what extent is it a discovery of a generic quality of intelligence, one that was going to work eventually, whenever somebody somewhere got around to figuring it out? If the latter, then AI is a lot less contingent, less sociomorphic, than it appears. Great minds are necessary to stitch the pieces, but eventually somebody was going to do it. Its inventors are less Promethean super-geniuses than just the people who happened to be there when some intrinsic aspect of intelligence was functionally demystified.

It does mean, however, that human intelligence is not what human intelligence thought it was all this time. It is both something we possess but which possesses us even more. It exists not just in individual brains, but even more so in the durable structures of communication between them, for example, in the form of language.

Like “life,” intelligence is modular, flexible and scalar, extending to the ingenious work of subcellular living machines and through the depths of evolutionary time. It also extends to much larger aggregations, of which each of us is a part, and also an instance. There is no reason to believe that the story would or should end with us; eschatology is useless. The evolution of intelligence does not peak with one terraforming species of nomadic primates.

Instead of people trying to write rules for the machine to apply to data, we give the data and the answers to the machine and it calculates the rules. This works tremendously well, and generalises far beyond images, but comes with the inherent limitation that such systems have no structural understanding of the question - they don’t necessarily have any concept of eyes or legs, let alone ‘cats’. 

If I ask for ‘the chest burster scheme in Alien as directed by Wes Anderson’ and get a 92% accurate output, no-one will complain that Sigourney Weaver had a different hair style. But if I ask for some JavaScript, or a contract, I might get a ‘98% accurate’ result that looks a lot like the JavaScript I asked for, but the 2% error might break the whole thing. To put this another way, some kinds of request don’t really have wrong answers, some can be roughly right, and some can only be precisely right or wrong, and cannot be ‘98% correct’.

Yahoo tried paying people to catalogue the entire web one site at a time, and that was unscalable. Google, on one side, is based on the patterns of aggregate human behaviour of the web, and on the other side it gives you ten results and makes you pick one - manual curation by billions of users. The index is made by machine, but the corpus it indexes is made by people and the results are chosen by people. In much the same way, generative networks, so far, rely on one side on patterns in things that people already created, and on the other on people having new ideas to type into the prompt and picking the ones that are good

But the other side of this is that ML gives you not infinite interns but one intern with super-human speed and memory - one intern who can listen to a billion calls and say ‘you know, after 300m calls, I noticed a pattern you didn’t know about…’