It is truly impossible to know what form the Technological Singularity will take. That, of course, is why it is called a singularity. However, several recent developments have crystallized in my mind how the world will look on the eve of the Technological Singularity. I think we can guess now what developments will be apparent days before it occurs.
The specific observations that have caused me to believe we can now see this clearly fall into two categories. The first category is observations in AI that look very much like intelligence but are not quite there. The most obvious example of this is the language prediction model GPT-3. Experimentation with this program has produced astonishing and disturbing results. Investigators have prompted it to argue that robots come in peace, and it produced the following essay:
https://www.theguardian.com/commentisfree/2020/sep/08/robot-wrote-this-article-gpt-3
I do not believe this is an exhibition of intelligence. In fact, it is the best argument I can produce that it is possible to know everything and understand nothing. I have met people like that, and I recognize it in a machine as well. Nevertheless, this has so many aspects of true intelligence that it greatly narrows the window for how the first genuinely intelligent machines will appear. People will be awestricken by their behavior, but still sense that there is something missing. The part that is missing will quickly fill in.
The second category of observations that have convinced me we can know when the Singularity is about to occur is instances where AI has helped to solve significant problems. A good example of AI doing this is the prediction of protein folding:
https://www.nature.com/articles/d41586-021-02025-4
As with the previously cited essay, this is more an achievement of brute force than of true intelligence. Nevertheless, the gap is narrowing. The fact that we have trouble discerning precisely where intelligence leaves off and this kind of behavior picks up is telling. What we call intelligence may rely more on brute force than we are prepared to accept.
With these two developments to draw from, I think we can reasonably ascertain what the world will look like on the eve of the Technological Singularity.
Probably, when that event is months, days or even minutes away, someone will develop a system, perhaps GPT-4, that astonishes people in a manner comparable to GPT-3. However, it will leave less room for skepticism. It may be asked to solve some real world problem such as designing a device that can consume ordinary regolith and turn out useful components. Such a device would be extraordinarily useful in the exploitation of asteroids. If we could send a machine to 16 Psyche that is able to do this, it could theoretically make the parts for more machines that could be used to mine 16 Psyche. The next step would be to ask GPT-4 to design a machine that could assemble the parts. If a system could actually do this, we could be assured that the Singularity is not far behind.
The aforementioned system would be put to work developing and creating all sorts of things, many of them dubious, until it is finally set to work designing yet another system that can do the job better. That advent is more or less the definition of the Technological Singularity and the thing we will really want to watch for. GPT-3 has been asked to design another language prediction model, which it did, but apparently not a very good one. GPT-4 may design a good one…possibly a better one than GPT-4.
On the day the Singularity is nearly upon us, there will be talk of manufacturing “humanoid robots” to replace human labor, but I suspect we will never reach that stage. Once the ability to outthink a person exists, in conjunction with the ability to create a humanoid robot, the next stages will come quickly. Whatever system is behind this proposed production will quickly imagine a better way of accomplishing the same thing and the Singularity will essentially be upon us. It is impossible to be sure what will be substituted for humanoid labor, but ideas are already in the works that strongly suggest there is a more efficient solution. There are proposals for robots that are very small and very large. A lights-out factory is essentially a robot with countless arms and other manipulative instruments. The humanoid robots will likely be deemed a useless intermediary step and fall by the wayside.
Other ways of reaching the Technological Singularity have been proposed in the past. These include human augmentation and the Internet “waking up”. However, developments with deep neural nets are progressing so much faster than human augmentation technologies, we can safely assume they are out of the running. If humans are ever augmented in any significant way, it will be for other purposes than advancing intelligence. The understanding of intelligence we have gained from neural nets also shows that the Internet simply lacks the structure to “wake up”. There is no way to get from where it is to what intelligence is. The Singularity will be reached through developments comparable to deep neural nets and especially GPT-3.
My intuition about where we are with programs like GPT-3 suggests that we will reach the Technological Singularity soon. We will certainly reach it by 2030, one of the earliest predicted dates for this event. No one born after 2010 will know what it means to have a career, earn a pension and ultimately retire in the age-old sense. In a way I feel sorry for them. It will be difficult to appreciate the world we are entering without fully appreciating the lifesways that brought us to this moment.
When it is clear that AI is being prompted for and proposing workable implementations that seem beyond what humans are presently developing, you can be sure that we are in the last stage of human evolution. I think it is fairly clear now what the world will look like on the day before the Technological Singularity. I haven’t any notion of what it will look like the day after.
spikosauropod 