Christie vs. Clinton: What Would Watson Say?
It's unlikely either Clinton or Christie will become president, so what's the point? It's to showcase how a system like Watson comes to a decision. It breaks the decision down into defined components: the quality of the data; effectiveness of the training; and scope of the analysis. While there is no doubt Watson could be used to predict who would win an election, a far better use would be to select who would make the better president.
01/20/14 5:00 AM PT
IBM is putting a massive amount of resources into Watson, which has the capability to dramatically improve the quality of our decisions. Since most don't know how Watson works, I thought it would be interesting to emulate how it might think when tied to a topical subject. As the U.S. approaches its next election circus, two candidates have jumped to the top as both frontrunners and targets.
Given how accurate predictions have been in the past at this phase of the election process, it's likely that one or both of these folks won't even make it to the primaries, let alone get elected. Still, I thought it might be interesting to try to emulate how IBM's Watson -- which would approach both candidates without party, sex, race or religious biases -- might choose if it weren't given a "none of the above" option. With both the Bridgegate and Benghazi scandals looming, which might be its favored choice, if it were given the option?
I'll close with my product of the week: the Windows Phone, the one phone that could save your life.
Watson is the first AI engine at a government-level scale. What this means is that it is designed to probe massive amounts of data to find the best answer -- not necessarily the right answer -- to a given problem. Like all data-based systems, including people, Watson's results can be corrupted by bad data and bad training. Yes -- part of how you build a Watson system is actually to train it over time so it can more accurately choose the right answer. It can be corrupted either by feeding it bad information or by having it trained by a biased or ill-informed trainer.
With accurate data and unbiased training, Watson likely could accurately call an election on the day of the election and would be more able to predict an outcome than any other computer-based system. That is because it can use massive amounts of data and make human-like decisions on each element to rank each one appropriately to form a conclusion. However, it is only as good as what it knows. It wouldn't know about events that hadn't occurred yet; however, it could infer and attach likelihoods to events that could alter the outcome of an election.
For instance, an untimely affair could be inferred by the personality type of either the candidate or the candidate's spouse, reducing the probability of that candidate winning. It also would look at things like the NSA leaks and could apply a probability to them and determine which candidate they would impact most adversely. In this instance, that would be Clinton, because she was in national office and more likely to be compromised by a discovery than Christie.
While there is no doubt Watson could be used to predict who would win an election -- my own sense is that Christie won't make it to the primaries and Clinton won't make it through them -- a far better use would be to select who would make the better president.
I'm going to make some assumptions because I'm not Watson. I don't have the ability to access nor the time to wade through the millions of bits of information it would have access to. Also, there is no training program that I'm aware of that would get Watson to the point where it could make this decision.
I'm going to assume Watson would have access to personnel files on both candidates and their public records in office. I'm also going to assume that its training would be by a centrist (like me) who doesn't align with either major party. I'm also going to assume that there are no big unknowns out there -- affairs, bribes, criminal activity -- that immediately would disqualify either candidate.
Finally, I'm going to assume Watson has access to medical data that would tell it how a candidate would respond to stress, as well as the likelihood a candidate would survive office or become infirm while in office.
Now I'm not Watson, and so my outcome is a guess. What I'm highlighting is how Watson likely would approach answering this question -- not the answer it would come up with.
How Watson Might Work
Training: This speaks to how capable the candidate would be in the office of president. Clinton clearly would be favored because she has been secretary of state and a senator. Assuming a normal management model, someone who has experience as an executive in an organization will be favored over someone who does not.
Performance: However, companies often hire from outside when performance falls below expectations, and both managed healthcare and Benghazi happened during Clinton's time in office. Further, the managed healthcare initiative appeared to originate with Clinton, and it has been a disaster so far. Christie has had a much more successful run in New Jersey and showcased an ability to cross the aisle to embrace the other party's president when it benefited his state's agenda. As a result, on performance, Christie would be favored. Objectively, performance would outrank training, because it more closely relates to a defined result.
Ethics: While there clearly are ethical issues with both candidates, Clinton would have a disadvantage because she has been in office longer, and she has been surrounded by a variety of scandals that would rank her down. While Christie comes from a state that would subjectively tarnish his ethics, objectively he just hasn't been tied to that many problems.
However, Bridgegate showcases a trending tendency to abuse power to take revenge, and with a nuclear arsenal and more examples, that almost assuredly would give Clinton the win here. As it currently stands, however, Christie likely would rank higher than Clinton. Watson likely would determine that the risk for a Christie catastrophe balances Clinton's history and call this a tie.
Execution: Here party practices would come into play, and this is where the Democratic practice of massive appointments of unqualified people likely would hurt Clinton. The more appointments of unqualified people, the more uncertainty is introduced, and uncertainty/incompetence can become a huge problem, as we have seen in the current administration. So Watson would tend to favor a Republican candidate because of predictability. Christie appears to stick with something longer than Clinton does, as well, and his accomplishments are better defined, favoring Christie.
Health: Clinton's age, particularly at the end of a second term, would create concerns -- as would Christie's weight and gender (men don't have the longevity of women). Clinton has had more health issues in public, due to stress, and Christie appears to be addressing his weight problem -- albeit not entirely successfully. Without access to either candidate's health records, and based only on what is public, Watson likely would conclude that Christie would be more likely to survive office.
Result: I think that were these criteria used, the result would be definitive: Christie, despite Bridgegate, would be Watson's selection as the better candidate.
Wrapping Up: So What?
I've already said I doubt either candidate will become president, so what's the point? The point is to showcase how a system like Watson comes to a decision. It breaks the decision down into defined components, and the result it provides is only as good as the quality of the data, effectiveness of the training, and scope of the analysis.
It is capable of being right far more than we or any other system, but only if all of the elements it uses to come to a conclusion are good. It suggests a process we can use to make better decisions: Break them down into components, and then even the most complex decision becomes far easier to resolve.
Product of the Week: Windows Phone
A few weeks back, I wrote about how I got beat up right in front of my house and how my Nokia 1020 let me down, because the fast picture feature had been disabled. I firmly believe that had I been able to get my phone into picture mode quickly, I could have prevented the attack.
Well, just before CES, I discovered that my phone had been patched. Once again, with the push of a single button and no PIN, I can take and send a picture to my SkyDrive. This one feature could make the difference between whether you live or die, and whether someone attacking you or a loved one might be caught.
If you can pull out your phone and shoot a picture of a criminal in the act of a crime -- or an accident -- that picture, or the act of taking it, could stop the event.
If I'd been able to take that picture on Halloween, I would still have the full use of my right arm, and I wouldn't have been on the ground unconscious with my wife screaming at the top of her lungs. I am convinced it would have stopped the attack -- and at the least, ensured my attackers would be caught. As a result, simply because this is the only camera with one-button instant picture capability, the Windows Phone is my product of the week. Thanks to Microsoft and Nokia for working to get this fixed!