Dealing with algorithm analysis, how much do you trust systems of artificial intelligence that are free from measurement bias in the decision making?
I am convinced that there has to be an intense analysis on what is the degree of trust in the results given by artificial intelligence. I remember an expression- I think that it was used by one of IBM leaders – that substituted the concept of Artificial Intelligence with Human Intelligence Artificially Assisted. At the end, we still have human intelligence resorting to a set of analysis produced by machines to take fast and efficient decisions.
There are those who think that machines can make isolated decisions.
Something that in part I agree with, if there are the conditions for it. One of them is that if there is a mistake, there has to be a protection that annuls or minimize the consequences of a bad decision. Most of all, there has to be a deep analysis on the degree of assertiveness of algorithms. Mainly on the benefits when they are right and the consequences when they are wrong. The work that must be done it’s mainly in the development of a process that evaluates the impact of false positives and true negatives in the algorithm results. That is, what we are considering as truth but that in fact it’s not, and on the other side, what to do when algorithms are telling that it’s not true but in fact it is. All of this can be mitigated if results can be analyzed in a second moment by someone that will make the final decision.
Can you explain it with an example?
There has been a lot of talk about promoting artificial intelligence in the diagnosis of illnesses. I think that no doctor will take a decision based on what is being said by only one model. What is expected is that there will always be someone that look at the results with the outmost attention and takes the final decision taking into account that individual case. In case there is the possibility that the answer’s given by the model are not right, then I believe that there will always be a second analysis in order to verify these results and to take the final decision of diagnosing or not an illness.
Autonomous driving has challenges that go beyond the analytical nature, there are also ethical: the classic dilemma of hitting an adult or a child. Are these questions still only at a theoretical standpoint or are people actually working on these issues?
There are working groups that are tackling these issues, by creating conditions so that machines can mimic human decisions. Although some of them are not unambiguous. Two people before the same situation would make different decisions. That will always be a problem, but I think that the main goal is to avoid collision as much as we can. Nevertheless, at a certain point, its possible that a situation like this is actually put forth and I don’t know how the machine is going to decide.
It’s still unknown.
It is, and when we talk about autonomous driving, there are several levels of assistance that have already been implemented, acting instead of people, managing the prevention of accidents. Mainly in those cases when people are not focused on driving. Many accidents happen because of a human flaw and the driving support systems performs a very important role in the reduction of the number of accidents.
We are talking about two different areas, but complementary to each other – one destined to improve the performance of the driver and the other in making more intelligent cars capable of isolated reaction. Autonomous driving seeks to optimize algorithms in order to guarantee that cars become more capable to reduce accidents. There is also the possibility of using artificial intelligence to track driver’s behaviors, detecting tiredness in the face, distracted or aggressive driving and so on. And also alert or teach people to be more responsible and to increase attention on the road.
With the expectation that road accidents will significantly decrease with autonomous driving, what future will the car insurance companies have?
Insurance companies do not see it as a bad thing the decrease of car accidents. They will obviously have a change in the type of products, on how the contracts are distributed, but this is already something that is happening today. All of these new types of mobility created with car sharing solutions are taking the emphasis off the private car, but that doesn’t mean that there are less cars circulating. The tendency will be to have more service providers, substituting private cars with company cars.
Will there be time to react to these changes?
Even if starting from today all new cars would be autonomous, there would still be a significant number of conventional cars, because our car usage has an average life span of 11-12 years. That is, this change will never happen so suddenly. Expectation is that, gradually, the automobile sector will end up by reducing its weight in terms of business volume. Probably there will be a change in the type of insurance sold – today the main one is of civil responsibility, but if the risks of accidents decreases, also prices will go down and clients will start purchasing coverage for personal damage.
One thing compensates the other
Exactly. And in a long-term thinking, there will be other entities that will have to assume their civil responsibility, mainly those who will conceive and manage the models and the actual circulation of automobiles. Other alternatives will appear, such as services associated to cyber risks, inherent to technology, or health, linked to the lifespan. It’s only natural that insurance companies evolve to adapt to changes. For example, 40 years ago, car insurance was not mandatory, cars where very few but there where already insurance companies that, with circulation risk assuming the most weight, grew in the automobile field. There was a time actually, when road accidents where catastrophic, although it was possible to reduce the number of accidents even if today there are much more cars circulating. Therefore, insurance companies are here to protect us and accompany all kind of risks that society, companies and people are exposed to.
What is your perspective on all of these future scenarios, many of them, very detailed?
My personal perspective is that the most extreme scenarios are not the most trustworthy, Technology advances very fast, but it doesn’t always becomes what we anticipated. There will always be unexpected changes. Many times, when there are very specific forecasts, reality overcomes them. When one is too categorical foreseen a specific situation developing in a very precise way in 5-10 years, generally, one misses the target. For example with the case of telematics in the automobile industry – that is, the capacity of monitoring and doing tariffs based on the specific driving of each person – a decade ago, it was the next big thing. After two, three years, a true explosion was anticipated that, after all, it still didn’t’ happen. And the traditional way of having your car insured is still there as one of the main components.
But the electrification of automobiles is more than a certain prediction.
Here, the role of regulation is fundamental. When there are specific dates imposed for electric vehicles to become compulsory – I think its’ difficult, for example, doing the same for autonomous vehicles – and if gradually regulation forbids the circulation of private cars in cities, that obviously powers the normalization of shared and electric cars, that gain weight and relevance in how we circulate in urban centers.