The United States has encountered a substantial number of problems this year — from the grounding of an entire line of Boeing planes to massive weather events, mass shootings, and a president who may be losing his bearings (seriously WTF?!?).
In each case, the focus often seems to be on finding fault rather than learning from and preventing the next disaster. Rather than investigating the causes of the problems to mitigate them, our leaders instead seem to focus on using the problems to help their political careers, showcasing a not uncommon but rather foolish emphasis on status over safety.
Last week I attended a briefing by IBM on its joint project with MIT to build a better AI, and it struck me that some of these same concepts could improve the quality of political decisions to shift attention more to fixing problems rather than using those problems as weapons against opponents.
I’m not talking about using artificial intelligence systems directly. I’m suggesting that we apply methods similar to those we now are considering for building and training them in order to make decisions that would fix major problems instead of just arguing about them forever. I’ll cherry-pick a few issues and explain how elements of AI training could be used to make better politicians and decision-makers.
As always, I’ll close with my product of the week: a laptop prototype from Nvidia that has the power to do AI modeling.
Brexit and Climate Change
I want to start by giving my thanks to England for making what we are dealing with in the U.S. seem so much less insane by comparison. One of the ways we are building AIs into things like self-driving cars is through massive simulations. We build virtual worlds and let the AIs experience them at machine speeds, and the accidents in these virtual worlds teach us how to improve the AI decision-making process without risk of real damage.
There are certain models of what is going to happen when the UK leaves the EU, and everything I’ve seen indicates it will be a train wreck. Because these models go against the will of the politicians who want Brexit, they are viewed as inaccurate.
The same thing is true of climate change. Most models indicate that we are looking at an impending disaster with existential potential, yet humans can’t come together to address what appear to be the core problems.
Much of what IBM has been working on is to create practices that result in unbiased AIs, which means identifying and eliminating the sources of bias early in the development process and making bias elimination a key part of the overall effort.
The benefit to an unbiased Brexit study likely could be informing voters in some detail of what they are in for — whether the eventual benefits are reasonably likely, and if they eventually would overcome the obvious massive adverse impact in the initial years.
It also would help ensure that the predictions of catastrophic impact would not become self-fulfilling prophecies, because the fear of Brexit alone likely could do — and is doing — much of the initial damage, as firms move out of the UK.
The nice thing about technology is that once a technology is built, you can scale it so that people can see the individual impacts of decisions. I imagine the UK voters would have been interested in a detailed personal impact report showcasing how Brexit was likely to impact them before they were locked in.
On global warming, given where you live, wouldn’t you like to know what the weather and living conditions will be like when you are older and perhaps no longer able to move with the weather? Or what your home will look like if there is an extreme weather event near you, coupled with a reliable probability of its occurrence?
At the very least, it might up your interest in getting or increasing your disaster insurance. Companies like Nvidia and IBM are building unbiased models that are predictive. That same approach could prevent nation-level mistakes if the outcomes could be trusted, and if people could accept results that countered their limited world view.
I’m convinced that neither U.S. political party wants to fix this problem. The Republicans want to be the defenders of the Second Amendment, although its interpretationhas changed so much over the years that it isn’t even recognizable. The Democrats want to deal with the symptoms and not the cause.
One of the big problems that the MIT/IBM team is overcoming is the AI’s ability to do causal analysis. One of the big problems we currently have in decision making with and without AIs is that correlation tends to drive decisions, but correlation doesn’t pinpoint causes. There is a massive correlation between mass shootings and guns, but there is virtually no evidence that guns are the cause of mass killings. They are the preferred tool in the U.S., but in other countries, other things are far more effective and available.
The example IBM used was one of margarine and divorce rates. While the graphs mirror each other over time, there has been no evidence of any connection, let alone causation.
The reason we should focus on the cause rather than the tool is that there are mass killings in areas of low gun use, and the people who commit them use alternatives like bombs, knives and cars to accomplish their goals. Explosives can result in a far higher death count and a far higher count of both critically injured and disabled people than an attack by a gun.
So, using worldwide data, we should be able to analyze whether getting rid of guns would result in fewer deaths, and whether there might be more easily accomplished alternatives that would massively reduce this problem. One possibility might be reaching an agreement with the NRA to curtail opposition to gun ownership in exchange for its agreement to stop blocking weapon safety advancements, to support better mandatory gun safety training, and to back a much more aggressive policy with respect to diagnosed mental illness, which would apply to all sellers.
On their current path, the NRA is going to be instrumental in getting the U.S. Second Amendment repealed or massively amended because, increasingly, it has been positioning itself directly on the critical path to preventing children from being shot.
The CDC could do the work to analyze what is likely a rich set of causes, but the Republicans have been blocking the effort, fearing the result obviously would be fewer guns. (By the way, given the CDC is independent, this would suggest it also believes it is on the wrong side of this.) However, the Democrats could fund this research, which would focus both on identifying the biggest causes and identifying practices that would lower the potential for mass killings.
I’m not a fan of the death penalty solution, because it appears that most mass shooters are prepared to die, and many do die in the process of carrying out their attacks. A more practical approach might be to allow a path to the death they seem to want without the interim step of killing a lot of kids. That would have to flow out of the research and modeling, however. In the end, much like IBM showcased with AI development, if you want to fix a problem, you need to do causal analysis.
If you don’t focus on the cause and limit yourself to examining the symptoms, just as in treating a disease, you have a high probability of making the actual problem worse. In carrying out the biggest domestic terrorist attack in the U.S., the attacker used fertilizer, not guns. Given the linked PBS program seems to suggest we have another of these coming, we don’t have a lot of time to identify and eliminate the likely causes for this scale of the attack.
Wrapping Up: Symbolic AI and Scientific Method
One of the concepts I found fascinating in the IBM/MIT talk was the use of a new type of AI. This type of AI, which joins machine learning and deep learning, is called “symbolic AI.” What makes it different is that rather than looking at an object, it looks at an object in component form, and then translates the components into language and finally builds a program from that language, which defines the object.
This form of AI is somewhat like the scientific method, which breaks down a complex problem into components that can be analyzed and solved more easily. Our political leaders would rather look right than be right — and that is seriously problematic. For instance, it seems that to avoid admitting he made a mistake, President Trump appears to have altered a weather map to make it look like he was right.
Here is the screwy thing: Not only was it pointless, providing false weather information is also a crime. We have the tools and practices to ensure that we make better decisions; it seems to me both foolish and life- and career-shortening to focus on appearing right when it’s obvious we weren’t.
Historically portable workstations either were desktop workstations with a handle that weighed a ton or laptops that were heavy, large and underperforming. Now the big problem is that these things not only pull a ton of power, but also generate a ton of heat.
If you were to build a thin workstation laptop, traditionally you’d either have to cut performance dramatically or risk cooking your privates if you ever put it on your lap, and neither is a very attractive outcome.
Well Nvidia, with its ACE Laptop Concept thought out this problem and put the components behind the screen, shifting the heat well away from your lap and up to where it would do far less harm.
It also used a unique Titanium cooling technology that allows the solution not only to dissipate the heat more effectively but also to add to the solution’s rigidity while keeping weight down.
The result is a five-pound class (what we once called “thin and light”) notebook with performance that ranges from slightly better than a similar desktop workstation to only about 10 percent worse depending on benchmark.
This laptop isn’t for everyone. It is for a select group of folks working on rendering, editing, development and even AIs, who need full-on workstation performance in a portable form. Price estimates for this thing exceed US$5K, so when this ships, it won’t be a cheap date either.
This innovative laptop was part of an impressive RTX product release at the IFA Show in Germany, and because it is an incredibly creative approach to a very difficult problem, the Nvidia ACE concept is my product of the week.