Moore's Law Pours Cold Water on Limitless AI Progress

The other day my wife read an article saying “AI that reads people’s emotions and makes recommendations is going to keep advancing, and so on,” and said, “that’s honestly terrifying.”
She used to be a nurse. Medicine is not omnipotent. Some patients can be cured and some can’t. Some patients are certain to recover but can’t stop worrying in the present moment, and some patients deteriorate before the test results even come back. In circumstances like these a nurse has to think seriously about what a person wants, what they’re thinking right now, and what kind of support she can offer, and then turn that into action. It’s an extraordinarily difficult profession.
What does he or she like, and what do they need right now.
I also talked before with a first-year doctoral student at Tohoku University. He researches cancer treatment using ultrasound, and he told me, “I don’t think curing disease is the most important thing a doctor should be doing.” Cancer treatment doesn’t end once the tumor is removed. What comes after — care not just for the patient but the family too, and, if it’s already too late, palliative care — runs very deep.
Analyzing needs and making recommendations. At first glance this looks like exactly the kind of territory AI is good at.
There’s a debate out there about whether AI could move into a domain like this. It will probably only heat up further from here.
But I don’t think AI can enter this domain. Here’s why. This isn’t an emotional argument of “I’m against AI!” or “we don’t need strong AI replacing humans!” It’s entirely a matter of logic. (For what it’s worth, I use AI constantly in my day job.)
Meanwhile there’s something that seems distant but is actually closely related to this: the “Moore’s Law” in the title. If you don’t know it, look it up on Wikipedia. In short, it’s the prediction that CPU processing performance doubles every 18 months, an exponential rate of progress.
It was around 2005, I think, when a very famous person asked on Twitter whether people agreed or disagreed with Moore’s Law. I disagreed. The reason is simple: silicon has physical limits, so process miniaturization can’t continue indefinitely. And indeed, the industry moved toward multi-core processing. In other words, it gave up on linearly increasing the speed of a single core and instead raised overall processing speed by making parallel processing more efficient. Put simply, Moore’s Law was wrong after all (which makes sense — I doubt even Moore himself expected his prediction to be quoted for this long).
As an aside, I once talked in Israel with someone from the team that first developed multi-core processors at Intel. At the time, plenty of people still believed in Moore’s Law, and headquarters in Santa Clara had apparently rejected the very idea of multi-core. The fact that they went ahead and secretly built it in Israel anyway struck me as both strong-willed and more than a little wild (meant as a compliment).
The idea that technology will keep advancing forever is an illusion.
Behind the scenes, engineers solve technical limitations through shifts in thinking, or swap in an entirely different solution underneath something that looks the same on the surface, and that’s how chasms get crossed. Underneath all that lies a graveyard piled high with the struggles, challenges, and disappointments of countless engineers.
The motivation to go that far only arises where there’s economic rationality — where something can become a business, or help someone in a way that can be monetized.
This is the basis for my own view that “AI that reads emotions and makes recommendations will not develop.”
Start with the technical hurdles. One is the generalization of vital sensing. For instance, if you think about ways to read emotion in real time, roughly speaking you get things like:
- camera (facial expression)
- heart rate, pulse
- body temperature
- voice (tone)
Needless to say, none of this is data lying around on the internet. It has to be captured on-site, which means some kind of device has to be developed. Most recommendation AI up to now has been based on behavioral data from the internet, and an enormous amount of it; compared to that, vital data gathered on-site is, frankly, an extremely weak database. It would be a serious mistake to think that what was possible in a world that lived entirely online can suddenly be done here too.
Now, suppose you somehow manage to develop the device. You still need to get users to wear it, or install it on-site, which means persuading them with an argument like “here are the benefits, so please wear this (or let us place this here).”
For reference, on the internet, when collecting information, there are prompts asking for permission to use cookies, and terms of service, and most users skim past these in a second even though they’re packed with rules. In other words, our behavioral data has already been harvested everywhere, for more than 20 years now.
Now, suppose you somehow manage to persuade someone, and data collection begins. Without data, AI accuracy doesn’t improve. And having personally worked with biological data many times, I can say from experience:
Biological data varies enormously between individuals, and it is genuinely brutal to deal with. That’s really the whole story. The time it takes to get stable results, the number of subjects needed, the ingenuity required in the algorithms — in short, the investment required. Unless you’re prepared for that, you can’t easily build an AI that works with biological data. This isn’t a question about AI technology itself, nor about supercomputer processing speed. It’s an entirely physical, mechanical, grueling saga demanding enormous labor and trial and error.
But that’s fine.
Human beings can push through hard things when there’s a purpose. The desire to grow is one of the levers of human self-actualization. Plenty of people pour their brief lives into clearing the challenge in front of them and standing on that summit. There will be plenty more to come.
So let’s reconfirm what that summit is.
“AI that reads emotions and makes recommendations.”
Wait — AI alone has no value by itself, so it needs to be tied to some kind of service.
So:
“A streaming service that plays music to match your mood that day.”
“A robot that comforts you when you’re feeling down.”
”
Hmm.
Does anyone really want that? At the very least, I can’t see anything here with the potential to be a game changer.
In other words, emotion AI is just another instance of the “illusion that technology will somehow keep advancing.”
Some people will research it for fun. But the amount invested in that will likely be less than a hundredth of what a “serious major corporation” would invest.
I think the key that made AI draw this much attention was the major progress in “image recognition technology.” Image recognition is extremely useful. It’s now used practically everywhere in real-world work. No need to explain further at this point. That’s because, starting with Google, there were an enormous number of “serious players” — not just companies but universities, research institutions, startups that sensed a business opportunity, and individuals with strong personal interest.
Emotion recognition hasn’t shown anywhere near that level of opportunity. At best it’s “might be interesting if it existed.” And that’s been true for a very long time now, unchanged. The lack of economic rationality and growth potential leads, mercilessly, to one prediction: this technological domain probably won’t advance at all.
Originally published in Japanese at https://clazytech.com/2020/01/419/. Translated with LLM assistance and reviewed before publication.