In a recent poll I shared, almost 70% of you said AI is not overhyped and are excited about its potential. This got me thinking: what will it take to fully realize this potential, and are we making the right tradeoffs as a society–as we move toward an AI-driven world? Let me delve into this further.
Today, society is incurring significant costs as we accelerate AI development. These costs fall into three main categories:
For example, did you know that the processing power needed for AI is doubling every 100 days? According to the World Economic Forum, by 2028, AI could use more power than Iceland did in 2021!
To make this a bit more concrete, training a large language (LLM) model like GPT-3 uses nearly 1,300-megawatt hours of electricity. That’s roughly equivalent to the annual power consumption of 130 U.S. homes, or 1.6 million hours watching Netflix, just to match the energy needed to train GPT-3! According to the Peterson Institute for International Economics, the carbon footprint of training an LLM is equivalent to about 626,000 pounds of carbon dioxide, leading to climate change and its many devastating effects.
The exponential growth of electricity needed at data centers (which is ultimately where AI models crunch away) also depletes our freshwater resources. For example, training GPT-3 in some U.S. data centers uses about 700,000 liters of clean water. Each time an AI model ‘thinks’ (a process called ‘Inference’), it again has to use computing resources. For example, ChatGPT answering about 20-50 questions is like the tool ‘drinking’ a 500ml bottle of water. As of June 2024, 627 million people use ChatGPT monthly–do the math, that’s a lot of water consumption! All this could negatively affect nature’s ecosystems and agriculture.
As of June 2024, 627 million people use ChatGPT monthly. At 500ml of water for every 20-50 questions, that’s a lot of water consumption.
We have all heard about Sam Altman’s ambition to raise $7 trillion from investors to build new factories to make chips specifically designed for AI, and about his plans to build a $100 billion data center with Microsoft for AI. Similarly, companies like Anthropic, Google Deepmind, Meta, etc., are all spending billions of dollars to train the latest and greatest AI models. This is a massive reallocation of financial resources from other societal/financial uses. Every time money is spent on something, there’s an ‘opportunity cost’ of what else that money could have been spent on. Imagine the positive social and financial impact those billions or trillions of dollars could have if they were invested in other areas.
According to a recent report from Goldman Sachs, AI could replace the equivalent of 300 million full-time jobs, and according to the McKinsey Global Institute, by 2030, at least 14% of employees globally could need to change their careers due to digitization, robotics, and AI advancements. While many new jobs will be created as people get reskilled, the disruption of employment (no matter how transitionary) caused by AI cannot be ignored.
In some ways, these three massive costs society is paying for AI are to be expected. As the saying goes, ‘no pain, no gain,’ right? Incurring costs for a major breakthrough isn’t necessarily futile, as long as they yield commensurate (or ideally, more) benefits than they cost. This is a standard way in which investors look at deals; they calculate a ‘return on investment’ (ROI). So the big question for us, as a society, is ‘Are we getting the ROI we deserve from AI?’
Right now, I would argue that the excitement and use cases being touted are largely NOT worth all the societal costs.
Much of the focus seems to be on chatbots and AI assistants that can do things like write poems, create songs, plan vacations or create meal plans for us. We’re thrilled with the ‘deep fakes’ and marvel at the ‘AI art’ that machines are creating. Startups with no revenue and no real use cases are getting billion-dollar valuations because they promise us yet another way in which AI will entertain or allow us to do our white-collar work with much less effort.
Don’t get me wrong; these feats are worth marveling at. Never before in history has technology advanced to a level of intelligence that is almost human-like. What I’m saying is, society should expect more from AI and hold it accountable. We need to see significantly better ROI from AI.
Anyone who knows me knows that I often say “Do Hard Things.” This comes from my belief in the responsibility that comes with privilege. The only way we can justify privilege is by using it to do really hard things for society, like solving big problems and creating opportunities for others.
Similarly, when we’re consuming all these resources for AI, we shouldn’t be doing so just for frivolous reasons. We should instead be using AI to solve the world’s most complex, difficult problems. I’m talking about truly meaningful things that unleash progress for humanity and don’t just make us lazier and entertain us. That’s the only way we can justify the environmental and financial “privileges” that AI is enjoying today.
Other articles that may interest you