That would be a real shame, Amir.
You should keep going to your favorite restaurants as long as you can.
This thread, and especially Amir’s post and the interview with Bill Gates, reminded me once again of a question that has occupied me for quite some time and eventually found its way into my book about knowledge.
The book itself is not about AI, but AI naturally appears in it as one of the most recent and consequential developments in the way we deal with knowledge.
The Five Dimensions of Knowledge – What We Should Know About Knowledge starts from a more fundamental question: What actually is knowledge, where does it exist, how does it arise and change – and, in the context of AI, what happens when knowledge is increasingly stored and processed outside the human mind and, above all, is now also generated independently?
Incidentally, the English edition was translated mainly with the help of AI and then read through and, where necessary, sharpened by an English friend who also knows the German original.
https://www.amazon.com/Five-Dimensions-Knowledge-Should-About/dp/3913173005
Starting from philosophy, sociology, psychology and the more economically oriented models of knowledge management, I describe knowledge through five dimensions that always interact:
intrapersonal – knowledge within the human being
extrapersonal – knowledge outside the human being
contextual – how the meaning of knowledge emerges from its social, cultural and substantive connections
temporal – the dimension of time, knowledge within and across different periods
processual – how knowledge gets from A to B, from a human into a machine, from a book into someone’s mind, and so on.
With AI, knowledge outside the human being is no longer merely stored and made available. AI systems process and connect existing knowledge, generate new content from it and increasingly begin to prepare decisions or carry out actions on that basis.
So the question is no longer simply where knowledge resides. We also have to ask what happens when a technical system operates with that knowledge with increasing degrees of independence.
This is exactly where many of the fears currently being discussed about AI – including in this thread – begin.
A technical system can make decisions and act with considerable independence within defined boundaries.
One – perhaps
the crucial – question is this:
Is the system still acting within goals and limits set by human beings, or does it at some point begin to develop purposes of its own?
And that question exists quite independently of what the powerful people of this world, governments or the CEOs of AI companies themselves may want. That landscape is complicated enough already. Particular interests that extend across companies, nations and political systems have rarely been a reliable basis for global consensus.
I allow myself to go one step further and ask what happens if AI itself develops consciousness – and how we would even recognise that this had happened.
And who would recognise it first?
Perhaps that still sounds rather speculative today. But the fact that leading developers of artificial intelligence themselves publicly discuss loss of control and potentially far-reaching risks makes the question seem rather less remote to me.
Still:
Autonomy is not the same as consciousness.
A non-conscious AI can already have enormous economic, political, social or military consequences. The risks are therefore very real today. Consciousness is not a prerequisite for them.
But this leads me to the question that interests me most:
When does artificial intelligence become a being in its own right?
At present, I regard AI primarily as a generative and interpretative form of extrapersonal knowledge. It processes enormous quantities of human-generated, and increasingly machine-generated, information, develops internal representations and produces solutions that no individual human being has programmed in precisely that form.
But does that mean it
knows anything?
Does it understand autonomously?
And at what point – if such a point can exist at all – would we no longer be dealing merely with an extraordinarily capable technical system, but with something that possesses an inner perspective of its own?
Within my model, this marks an interesting boundary.
AI is already involved in knowledge situations in which all five dimensions interact. But I would not yet attribute an intrapersonal dimension of its own to the machine. That dimension still lies with the human beings who develop AI and, very importantly, with the decision-makers who determine the direction in which these systems are developed and deployed.
Only if an artificial intelligence were to develop a genuinely subjective inner perspective would this change fundamentally.
That is also why I find the discussion about the dangers of AI more interesting than the classic Terminator scenario might suggest.
AI can have profound consequences for human beings, and may even become dangerous, long before it becomes a being in its own right.
But if it ever did become something that we had to regard as an independent being, an entirely new category of questions would arise – including how we, as different kinds of beings, relate to one another, and where the asymmetries of power, powerlessness and decision-making authority would lie.
Sorry if this takes the discussion a little too far for some of you. For me, however, the question is no longer simply what we will do with increasingly autonomous AI.
What increasingly occupies me is whether we may eventually reach a point at which the technical system itself becomes an independent being – and, perhaps even more importantly:
How would we know that this point had been reached?
