Artificial intelligence is unquestionably the most important technology to be developed since atomic power.
When the Atomic Age began in 1945, nuclear technology offered us the promise limitless power. Unfortunately, nuclear technology carried with it existential risks of nuclear destruction.
Today, the AI Age is just beginning, and artificial intelligence offers us the promise of limitless processing power, of limitless intelligence even. Unfortunately, many AI experts see existential risks from AI, too.
With hindsight we can see that nuclear technology failed to live up to its promises but more than delivered on its risk. Far from enjoying limitless power, the world today remains addicted to fossil fuels. The few reactors that do exist are slowly being phased out due to the high cost of fuel and the dangers of meltdown and fallout. Unwise decisions made decades ago have ruined the promise of limitless energy while worsening the existential risks through weapons proliferation.
As pioneers in AI technology, we have to ask: What went wrong in the Atomic Age? And how can we avoid making the same mistakes with our technology that the atomic pioneers made with theirs? How can we steer AI technology to fulfilling its promises while minimizing its risks?
To answer those questions, we need to travel in time back to the Atomic Age, back to the futuristic year 1973. Five years earlier, in 1968, the discover of plutonium, Nobel laureate Glenn Seaborg, had told the Atomic Energy Commission that the Oak Ridge National Laboratory had successfully developed and tested a thorium-fueled reactor. Thorium-fueled power, he explained, had a number of advantages over the uranium-fueled reactors then in use.
- One ton of thorium can produce as much power as 200 tons of uranium or 3,500,000 tons of coal.
- Thorium is three times as abundant as uranium – with enough thorium in the U.S. to power our country for a thousand years. Mining that thorium is both safer and cheaper than mining uranium, and also less environmentally damaging.
- Reactors using thorium create between 100 and 1000 times less nuclear waste than reactors using uranium, and the radioactivity of their nuclear waste drops to safe levels within a few hundred years. Uranium waste can take tens of thousands of years to cool off.
- Thorium-fueled reactors have little to no chance of nuclear meltdown.
- Finally, thorium-fueled reactors produce 98% less plutonium byproduct than uranium-fueled reactors, making militarization of the byproducts into nuclear weapons impractical and difficult.
Despite these benefits, the US government decided in 1973 to make uranium, rather than thorium, the center of its atomic program. Alvin Weinberg, the director of the Oak Ridge National Laboratory, was forced out and his thorium reactor was shut down. Science writer Richard Martin explains:
Weinberg realized that you could use thorium in an entirely new kind of reactor, one that would have zero risk of meltdown. … his team built a working reactor … and he spent the rest of his 18-year tenure trying to make thorium the heart of the nation’s atomic power effort. He failed. Uranium reactors had already been established, and Hyman Rickover, de facto head of the US nuclear program, wanted the plutonium from uranium-powered nuclear plants to make bombs. Increasingly shunted aside, Weinberg was finally forced out in 1973.
The U.S. had the opportunity to invest in a nuclear power source that was abundant, easy to mine, safe to use, and impossible to weaponize. It chose the more dangerous, more difficult to mine, and less safe to use uranium technology because it was easier to weaponize.
So complete is uranium’s dominance that today, many PhDs in nuclear reactor technology do not even know about thorium. it’s possible to have a Ph.D. in nuclear reactor technology and not know about thorium energy. For instance, nuclear physicist Victor J. Stenger in 2012 said, “It came as a surprise to me to learn recently that such an alternative has been available to us since World War II, but not pursued because it lacked weapons applications.” Meanwhile, NASA scientist and thorium expert Kirk Sorensen has called thorium “the alternative path that was not taken”. If the US had continued its research in 1974, he believes, today America would have total energy independence.
But we chose uranium instead of thorium, and so we never got the limitless power we might have. Instead we got Three Mile Island, Chernobyl, Fukushima. We got mutual assured destruction. We got aging uranium power plants that produce toxic waste that lasts forever, and which are being shut down worldwide by anti-nuclear Green parties who fear the dangers they pose.
At this moment, in 2022, AI’s early adopters stand at a similar juncture. They have a choice between two technologies. On the one hand, they can choose neural network technology. Like uranium-fueled reactors in 1973, neural network technology has been around for decades. It is well-established. And it is unsafe – its core a black box of opaque decision-making, its choices as inauditable as they are inexplicable. They are impossible to align with the intent of their users because their users have no idea why they do what they do.
On the other hand, they can choose instance-based learning technology. Like thorium-fueled reactors, instance-based learning technology has been around for decades, but it has been overlooked even by experts in the fields. It is not yet well-established. But it is far safer than neural network tech, because it is auditable and explainable – it is what we call Understandable AI™. And, like thorium-fueled reactors, instance-based learning technology isn’t just safer, it’s actually better than its rivals.
If the AI industry takes the path of neural networks, it’ll be making the same mistake the atomic pioneers made, choosing the more dangerous technology simply because it’s more established. And, we fear, a world that widely adopts neural networks for mission-critical services will likely be a world where anti-AI activists come to power and shut down the technology when it proves uncontrollable and risky.
At Diveplane, we call our instance-based learning platform Diveplane Reactor. We don’t intend to let our Reactor be on the path not taken. The future deserves better, and we all deserve a better future. Understandable AI® will help us get there.