The previous seven blogs in this series have covered quite a lot of ground on this very multi-faceted topic of AI. There is a lot of uncertainty as to how AI will alter the way the world looks economically, environmentally, geo-politically, legislatively and socially. There is also a lot of uncertainty as to the technology AI will bring us in the future, as well as how the technology of AI itself will progress. Naturally, it is easy to get overwhelmed with all the information and all the ways in which things could go wrong.
Overall, I would say that drastic change is not coming in the immediate future, by which I mean the next few years. On the other hand, it’s not a million miles away, and I do expect the world to look noticeably different in five years, very different in ten years, and extremely different in twenty years. Speed of progress in all fields is about to reach heights never even dreamt before, but in opposition is the time it takes for the required infrastructure to be implemented for widespread AI use. Likewise, it takes time for corporations and large populations to adjust to such drastic changes.
At the same time, a lot of this expectation relies on AI progress continuing over the same exponential curve it has been. While there is no reason to believe progress will slow down for now, several top voices have been saying we are moving from an era of scaling to an era of research, which I would agree with. By scaling, they mean improving LLMs by scaling it on more data. As of now, this is a philosophy which is starting to see diminishing returns. Reinforcement learning, which until recently was seeing good returns in AI progress, is also starting to slow down in progress as well as creating certain potential issues, leading to questions as to the path to AGI. Thus, these companies are forced to look elsewhere, which is the exact goal of research. Who knows, maybe we hit a wall at this stage (again, I think fully stagnating is unlikely). Likewise, maybe a huge innovation comes through, even if not immediately, which speeds us up beyond the extrapolated exponential curve we are hoping for AI progress to follow.
Likewise, while the economy is not going to be absolutely crushed in the next few years, we are seeing layoffs even now. It’s really hard to predict how quickly new jobs will pop up in the space of old jobs which have disappeared, and very potentially the start of this replacement will coincide with the peak of unemployment (following from basic logic). How long this replacement takes will likely dictate how bad unemployment gets before any sort of self-correction. Again, I do not think this will be immediately dystopian, though on an individual basis this could cause a lot of trouble for those who end up unemployed. There have also been talks of “AI washing” where companies blame AI for layoffs as an excuse, when the motivation was actually for other reasons such as over hiring or any other reason which the company would want to hide from the public. I saw a YouTube video of a former Amazon employee who said she could see it coming from a mile away and had quit ahead of the mass firings. She made claims of ridiculous amounts of employees being given tasks that need 1/3 the number of people to do consistently alongside extreme mass hiring as well as the potential motivation of seeing the increase in employment size as representing the company expanding their chips, without thinking logistically on what the incredibly large number of employees could provide. When Amazon claimed they were firing due to AI, she claimed it was clearly due to grossly over-hiring.
This account points to an interesting thought that this mass firing issue could potentially be not as bad as we think, depending on how many companies are doing this. My gut instinct is, even if it is a factor, it is likely there are genuinely a big rise in cases of layoffs as a direct result of mass AI adoption. It’s a tough time to graduate, but for broader society I don’t think fear needs to be coming in quite yet and there is time to try upskilling yourself and educate yourself in a way that better positions yourself for the future if you are in a role which is at risk (non-managerial and repetitive grunt work jobs, particularly coding).
While it is usually overestimated how much AI negatively affects the environment, the ambitions of exponential growth of power usage to train these models could tip the scales towards AI being notably harmful to the environment. Regardless, for all we know, with the expected explosion of scientific discovery expected to happen (with demonstrable accelerations already taking place at AI infancy), we could make huge strides in reverting global warming with the help of AI. Likewise, there are also plans of Elon Musk and google trying to send these chips to space, circumventing this negative as well as the negative of hogging up local electricity supplies. Water cooling techniques to cool the chips used to train AI has also been a concern for hogging drinking water. I read a news article about the Chinese using the oceans the cool these clusters of GPUs, a very interesting idea! This could prove to be an alternative which does not harm local communities.
While we do not know of a lot of the different technology we will see come up down the line as a result of AI, there are a lot of visions being implanted into our minds (not always figuratively, like potentially with Neuralink!) Regardless of the specific technology that we will see, what we do know is there is immense potential for efficiency gains never seen before. From bringing back movement in paralyzed bodies, to driverless cars, to accelerated drug research and scientific research in general, to robot assistants (that is, with a physical body) who can rapidly do labour 24/7, to agents doing grunt work automatically, the list of potential uses are huge. Generally speaking, the public overestimate the range and capabilities as well as its effects on the world in the short term (few years), but underestimate what it may do in the long term (10+ years).
Given the unimaginably vast number of scenarios these chatbots are put in when at the hands of the general public and all the edge cases (and even non-edge cases!) where consumer detriment is possible or even probable, as well as the issue of making sure these chatbots stay aligned with human interests makes legislation and regulation incredibly important to prevent AI induced harm to general society. With the speed of AI progression, this will really test the speed of the courtrooms, particularly in bureaucracy heavy countries. And this is just to get any legislation out, with how new this technology is, we have little information to go off, and the courtrooms will largely be shooting in the dark, even if with the help of experts. Who knows how long any potentially harmful effects of AI will live among society until sufficient legislation is in place. Particularly in countries with less competent and/or corrupt governments!
The way the public views depends as much on the information fed to them as it does genuine tangible shifts seen contemporarily. Unfortunately, the need for profit incentivizes divisive and polarizing content over accurate and informative content, and this in my view negatively affects perceptions of AI, much like many other topics that this unfortunate mechanism plagues. Optimism varies largely among different countries with different countries and narratives circulating, with Asia generally being a lot more positive than the west. There is also a big risk of people losing their ability to think critically and miss out on a lot of learning by leaning on AI as a crutch rather than a tool to help them. Same can be said with emotional regulation and needs. AI partners as a product have a huge market now, and this is likely to only grow, which could cause a lot of problems!
Overall, to repeat myself a bit (because it is worth re-iterating), I do not think the world is ending in the next few years, though I do think it will be transformed within the next decade or two. I believe this to be true over the many different angles of this nuanced topic. The main thing is to control what you can control rather than feel depressed about what you cannot control. In my opinion, this is to try and get in the best position possible in time for the very realistic scenario of “shit hitting the fan”, in my view moderately far down the line.