on
AI and the coming inequality
A lot of focus thus far on AI has been on safety. And rightly so. Safety of any new technology is critical before it should be made available to mass markets.
But how about societal inequality? Inequality today is quite stark. The wealthy 1.5% own 48% of the world’s wealth. The fundamental premise of AI is to boost productivity. What happens when rich nations / corporations (who also happen to own the means of producing these models) get boosted productivity by AI ? AI enabling tech will then become a dependency for every product. The rich get richer and monopolistic. Everyone else gets improved productivity but not at the same scale. Resulting in worse inequality.
It’s different this time
Innovation has always bred some discontent. The industrial revolution, steam boats etc made many trades redundant. That did not stop any of these innovations from going mass market.
The fundamental difference though is that the means of production were readily reproducible everywhere. The printing press, assembly lines etc could be set up in cheaper parts of the world to 10x production. Which is exactly what happened. Asia became the factory of the world producing goods and services, thereby getting large swathes of the population out of poverty.
One could argue that monopoly has been the case for a while. TSMC manufactures 90% of the world’s most advanced chips. ASML has a similar market share for photolithography machines. The difference though is that the end product that incorporated these chips were manufactured by many different vendors. There are hundreds of manufacturers for vehicles, cell phones, washing machines etc that utilize these chips. That lets the prosperity be distributed.
With AI however, the core IP is knowledge. All you need is a bunch of ML engineers who can suddenly render an entire profession / use-case redundant. And only the final product, perhaps sent over the wire, is needed at the other end. I.e. “the means of production” / intelligence will be owned by fewer and fewer people / corporations / countries. As the use of AI proliferates, the differentiator between products will become the access to these AI productivity enhancing capabilities, all of which would be owned by fewer and fewer corporations.
The “secret sauce” of future products will become their AI capabilities. You will not have much of a moat if you outsource that.
More power, concentrated in the hands of fewer and fewer.
Winner Winners takes all most
AI has some strange dynamics. The more you use it, the better it gets. Take self driving for example. The clear leader in this space is Waymo. Waymo has 20M+ miles of real world driving data. That is more than their closest competitor by quite a lot. With every new mile added, the AI models learn more and just get better learning from different real world scenarios. It is already better than humans in terms of fatalities and accidents.
This is hard to replicate. Others will always be playing catch up. The next time you want to order a self driving taxi, the option people choose (from an already slim number of options) will be the one which is the safest i.e. one that has already done the most miles of flawless driving.
It’s hard for established car manufacturers to jump in too. It’s not that Toyota or Honda can decide that next year they want to get into the self-driving race too. It takes years to build such capability, so they will have to partner with some OEMs to get this technology.
The winners will be winning a whole lot more.
AI Haves and AI Have Nots
AI is a self reinforcing paradigm. The more data you collect, the better your AI gets, the better your services get which help you corner a larger share of the market. There are challenges in terms of talent as well as cost that are not easily solvable.
In the past, for similar technologies, governments stepped in when a technology was deemed important to national security. E.g. Nuclear. Governments pumped in resources so as not to be left behind. But that has not happened wrt investments in AI. Why ? One reason may be that the use cases for AI are still up in the air. Productivity is a sure win, but again, that’s hard to define. So governments have been slow in recognizing the need for investments in AI. Private enterprise has taken the lead and most of them are in the US or China.
This gives corporations tremendous leverage in improving existing products and also in attacking problems at the fringes of our reach thus far. Consequently these new capabilities / discoveries will also be owned and operated by few.
The cost and intellectual capital needed to make these models is also non-trivial. Out of reach of most. Compare this to previous waves of innovation e.g. the mobile revolution. With mass production the cost of handsets came down drastically. You also did not need specialized knowledge or mountains of cash to build out apps, marketplaces etc Owing to which they proliferated everywhere and the benefits of it were usable by all.
That’s not the case with AI. The low hanging fruit of AI has already been taken. Every new leap forward will require much more heavy lifting, much bigger/specialized clusters…The xAI cluster costs around $3-4B, Meta is spending billions on it’s clusters as well
Less competition, more monopoly.
Geopolitical Ramifications
Right now the center of gravity for all AI related development is around the US and China (with Europe making an occasional guest appearance). These two also happen to be the two superpowers embroiled in tensions.
Already AI is finding its way into military applications. Conventional militaries are being replaced by drones, autonomous vehicles and so on. Computer vision has significantly advanced. What would take analysts pouring over images / videos to detect events of interest can be done via AI in a fraction of the time. If the war in Ukraine has taught the world anything, it’s that your expensive defensive shields can be overwhelmed quickly via cheap and intelligent drones. Today’s wars have become testing beds for future wars. And AI will be the main differentiator.
And so nations will soon have to choose alliances they will be part of. For survival. Make a military alliance with one and get shut out from any economic cooperation with the other.
With only a few prominent AI players, the world will get carved out between them. Stark similarities with the 1800s when the world got divided into regions of influence by the dominant naval powers.
More power, fewer players.
What should we do ?
Open Source
Open source everything. It opens the floodgates of innovation. Other companies that are not open sourcing their latest models, could however open source the older versions. For many vertical industries using just those older versions would help in significant productivity improvements.
The big players cannot and don’t want to be in every vertical. Open sourcing tools, techniques and processes related to the latest in AI, can kickstart development by smaller players thereby spreading the gains of AI across society.
Government Action
Governments need to recognize the potential and dangers of missing the AI boat and start funding initiatives to improve access to AI. e.g maybe fund new GPU clusters that can be used for training or investments in companies for AI research. Identify all monopolies and create funds to make alternatives.
There has been lot’s of talk about regulating AI. I don’t think they will work. (except in military applications)
If there is one thing governments have, that’s money. Utilize it wisely.
Treaties to prevent AI in use of war
What stopped nuclear arms from proliferating was mainly the international treaties. The same needs to be done for the use of AI in lethal weapons. Nations need to pass regulation to prevent use of AI in lethal weapons. (In the past advanced technologies with deep ethical concerns like germline gene editing have been banned)
This requires getting nations together and understanding the risks of AI in war and actively passing global binding resolutions to ban it. A tall order indeed.
Finally…
Things move fast. The AI takeover will be gradual and then all of a sudden. To be prepared, we need to start thinking now. How will our new found productivity impact economies? We have already learnt that trickle down economics does not work. How do we ensure the benefits of a once in a generation technology is equitably distributed ? This may ultimately turn out to be the toughest challenge facing AI thus far.