In Defense of AI

It's not just chatbots and video slop

Sometimes I worry that I’m giving the wrong impression around here.

I write quite a lot about many of the negative side effects of technology, including what’s going on with social media and AI. I think Posting Is The Most Powerful Force In The Universe, and not in a good way. I worry about influencers, about gambling, about what technology is doing to the social contract. I yell at venture capitalists. I talk about how smartphones and social media are causing all kinds of harmful social phenomenon and how technology profanes the sacred.

You could be forgiven for thinking this is a tech-skeptical blog, one that only exists to promote the worldview that technology is harmful and needs to be tightly controlled. This is especially true when it comes to AI - I’ve written about AI video slop, AI chatbot psychosis, and the many ways AI content is taking over social media. But the truth is that at heart, I’m a techno-optimist. I believe in the transformative power of technology, and that includes AI. So I want to explain why so much of the blog focuses on the negative, and why despite all the negative aspects of AI it’s going to have a transformative impact on society.

---

Part of the reason that this blog focuses on the negative so often is that it’s easy to write negative stories. Something that’s new has an unexpected side effect, people get upset, you summarize what happens, make a few pithy observations, and Hey Presto you’ve got a story. Negativity bias is everywhere in the media - local TV stations and newspapers worked for decades off the informal motto “If it bleeds, it leads”.

But beyond simple negativity bias, part of the reason I’m always talking about the downsides of AI is that social media is dominated by two particular use cases of AI: LLM chatbots and video/image generation. These two modes of AI are ubiquitous online, and can be incredibly controversial. If you’re mad about AI it’s almost certainly one of those two things. Concerns about copyrighted images? Fake AI-generated images of a politician? Slop content dominating your YouTube algorithm? Deep faked non-consensual pornography? All due to AI-generated videos and images. Meanwhile AI chatbots are the primary concern when it comes to social media bots, AI psychosis, and more. I share some of these frustrations - I’ve been stunned by how real-world humans seem to be outsourcing their common sense to chatbots:

This really happened, and it was incredibly bizarre.

But here’s the core of my pitch. Even if you hate AI art, AI video, and LLMs, you should still be optimistic about AI in the long run. There’s so much more to AI than just those things. AI is helping people right now, today, and it has the potential to help billions in the future in very tangible, non-digital ways.¹

How is AI helping people? I’m glad you asked.

Self-Driving Cars

In America, around 40,000 people die every year from car crashes. Millions more are injured in car crashes every year. And the evidence is very strong that driverless cars are much safer than human drivers:

Waymo estimates that typical human drivers would have gotten into airbag-triggering crashes 159 times over 96 million miles in the cities where Waymo operates. Waymo, in contrast, got into only 34 airbag crashes—a 79 percent reduction.

In fact, that 79% reduction in serious crashes almost certainly downplays how much safer Waymos are compared to conventional human drivers. When the blog UnderstandingAI actually went through every NHTSA report for a Waymo accident, the vast majority of airbag crashes were other drivers running into Waymos - often while the Waymo was completely stationary. Only in a tiny minority of crashes was the self-driving software even partially at fault.

We have every reason to believe that the coming transition to self-driving cars will save tens of thousands of lives per year, as well as preventing millions of injuries. It will also revolutionize many aspects of everyday life. We’ll need less space in our cities for parking, freeing that space to be used for retail, housing, parks, etc. People will be able to read, work, sleep or play during their commutes. Police will be able to re-allocate their time to solving serious crimes rather than endlessly monitoring traffic. And this transition is already happening - you can take Waymos in multiple cities right now, today. They’re expanding and will soon start driving on freeways. Driverless long-haul trucks are coming soon as well. It’s a big deal, and something that we should all be excited about.²

AI-Boosted Manufacturing

The best way to show how AI can be used to improve manufacturing is to just show an example, like how auto manufacturer GM re-engineered their seat brackets using AI-powered ‘generative design’:

The benefit? The end result is structurally stronger while using less raw material and having fewer parts. It also looks like something from an alien spacecraft, which rocks.

This sort of AI-assisted design is already producing improved parts in our cars, rockets, planes, and in all kinds of consumer goods. And that’s just one way AI is being used - it can also help redesign plant processes, help iterate prototypes, speed up simulation and testing, and more. It’s making the things we buy better and more efficient bit by bit.

AI in Medical Imaging

A wide variety of medical imaging is currently being automated using AI techniques. AI currently either matches or outperforms medical professionals when it comes to detecting breast cancer, interpreting echocardiograph results, detecting lung nodules, finding melanomas, diagnosing tuberculosis and more. By looking at images of the brain, custom AI models are able to diagnose patients with Alzheimer’s up to seven years before they show any symptoms. Custom models also show great promise in predicting Parkinson’s disease from voice recordings.

The benefits here are potentially huge. AI can catch more diseases early, saving lives. AI also reduces the workload on radiologists - and contrary to dire predictions, most countries are experiencing a shortage of qualified radiologists, so this is a big deal. Less time spent poring over blurry images means more time for radiologists to do everything else - building plans for care, communicating with patients and other doctors, etc. There’s also evidence that AI can help doctors and patients make better diagnoses, especially tricky diagnoses.

Instant, Free Translation

In the Hitchhiker’s Guide to the Galaxy, our hero meets a bunch of aliens and is able to understand their language by inserting a Babel fish into his ear. The fish has special properties that allow the user to hear any language as their native language, and it was necessary as a plot device to get around the question of “Why do all aliens speak the Queen’s English?”. It obviously wasn’t realistic, but in science fiction novels it’s normal to invent crazy new forms of technology.

Just a few decades later, Babel fish are essentially real.

I worry that most people don’t fully appreciate how remarkable this is. It is absolutely insane that I can go to virtually any place in the world, whip out my phone, and instantly be able to translate any text in any language. If I need to talk with locals, I can speak into my phone in English and have the words instantly translated, written or spoken, into any language I choose. And I’m describing it the old-fashioned way. Apple now offers real-time live translation in current gen AirPods. It’s almost exactly the same thing as having a Babel fish in your ear! Imagine telling someone a hundred years ago that translation into any language would be universal, instant, and free. It’s astonishing.

Weather Forecasting

Weather forecasting is undergoing a revolution:

The new A.I. forecasts are, by leaps and bounds, easier, faster, and cheaper to produce than the non-A.I. variety, using 1,000 times less computational energy. And, in most cases, these A.I. forecasts, powered by machine learning, are more accurate, too. “Right now the machine learning model is producing better scores,” says Peter Dueben, a model developer at ECMWF in Bonn, who helped to develop the center’s Artificial Intelligence Forecasting System (AIFS). The improvement is hard to quantify, but the ECMWF says that for some weather phenomena, the AIFS is 20 percent better than its state-of-the-art physics-based models.

What sort of real world impacts is this having? AI is improving monsoon season rainfall predictions in South Asia, giving more accurate forecasts further in advance, which has large benefits for farmers. AI is also better at predicting the paths of hurricanes than conventional models:

Contra to the usual concerns about AI power usage, these AI models are also faster and cheaper to run than conventional weather models.

—-

Frankly, the examples above are just a few of the biggest examples of AI that impacts our everyday life. There are so many more stories out there:

• AI is being used to more quickly identify potholes in cities, leading to better street maintenance.

• AI is being used to identify illegal fishing activity and protect wild fish populations.

• AI has completely, utterly solved the previously intractable problem of protein folding.

• Machine-learned force fields could make it orders of magnitude faster/easier to model and design new materials.

• AI-targeted pesticide spraying tools can precisely target weeds and reduce the use of pesticides by 70%.

• AI models can predict the aftershocks of earthquakes better than conventional models.

If you want to peruse dozens more examples, you can check out the site AI Opportunities.

---

What you may notice here is that virtually none of the positive uses of AI listed above have anything to do with chatbots or image/video generation. Most of these uses are in professional settings. There’s genuinely so much that can be accomplished by applying AI to hard problems.

It frustrates me endlessly that so much energy, time, and brainpower is instead devoted to large language model chatbots and AI video. I don’t think LLMs are useless! I use them frequently. But I do think that we should divert at least a little bit of the energy going into chatbots and instead put that energy towards advances in the hard sciences, in predicting disasters, in diagnosing and preventing diseases.

There are signs we may be entering a new age of cross-partisan anti-AI politics. So my plea for you all is this - even if you think that ChatGPT is the devil, don’t toss out the baby with the bathwater. Let’s keep pushing to accomplish tangible things with AI, even if we worry about the social consequences of chatbots while we do it.

1. Fair warning - the things mentioned below use a mixture of different concepts like machine learning, transformers, reinforcement learning, neural networks, etc. They’re not all the same, but I’m going to simplify by calling them all ‘AI’, which might be wrong in a nitpicky technical sense but not in how we understand the concept of AI as a society.

2. Some folks worry about the job loss aspect of driverless cars, and it will be a shock to our economy if millions of people lose their jobs all at once. But the transition is almost certain to be gradual, and this kind of change is ultimately healthy. It’s good that we no longer have milkmen, that we no longer have elevator operators. Creative destruction is why we’re richer than our grandparents.