Trusting the Digital Assistants

There are things that are nice and interesting to do, yet there are things you’d rather not spend your time and effort doing. People are different, and my categorization may not match yours, but on average, there are a lot of things in today’s life that one would like to delegate to some extent. Hiring a person or a team for this is something very few of us can afford, but technical progress gives some hope to wide audience, too.

Of course, we don’t each get our own human-like robot Jeeves tomorrow, but outside of the physical reality, in the information space, the so-called “digital assistants” do enter our lives. They do so very slowly, and there’s an important reason for that. In fact, our “digital assistants” are right now trying to reach a milestone they’ve been having a hard time getting to for a while, not entirely for techincal considerations.

The milestone is the fundamental difference between “OK, Google, show me the flights to Moscow for Friday” and “hey, Siri, book me a Friday ticket to Moscow, please”. The assistant giving you the information it collected is one matter, the assistant making a decision is a whole different deal. Delegating decision making is not easy, even if we speak about a human assistant: you need to be sure that not only the assistant is capable of performing the task, but that the choices will be done to your benefit.

For a human assistant, there must be a strong motivation to act in your interests. How that can be achieved is the subject of thousands of books: financial incentives, personal loyalty, religious motives, morals, etc., etc. A digital assistant only needs to be programmed for the desirable behaviour (that is, to optimize for the parameters you are interested in). There’s only one issue: you are not the one programming the assistant.

It’s nice if the decision can be easily verified. If your phone camera “misses” on the white balance, you notice immediately. Decisions like this one have been delegated for a long time now, no issues there. It’s also nice when the interests of whoever made the assistant match those of the user: people who program spam filters usually sincerely want your inbox to have less spam. In this case, however, a conflict of interests is possible: may not be very incentivized to filter spam coming from VK, and Yahoo mail may want to give Verizon spam some slack, for instance…

Unfortunately, as of today, only big companies have means and expertise to develop more general-purpose “digital assistants”, and the conflict of interests is practically unavoidable. Choosing betweed you and Apple Inc., to whose benefit Siri in your phone will act? What are the chances Alexa chooses to make a purchase from Walmart instead of Amazon, even if it’s cheaper and the delivery is faster? People are different, of course, but I seriously doubt anybody can trust the assistant with making decisions, even if this person is “all in” on the business-ecosystem in question.

The logic suggests that this conflict of interests can only be avoided if the developer of the “digital assistant” has no other businesses and is not affilated with other companies. Naturally, the service would require subscription, and the price will be steep, at least initially, but as far as I understand, right now no reasonable subscription price would cover the development of such a system. We can only hope the further technical progress will make the development of a digital assistant cheaper. This model, however, doesn’t solve the issues of privacy and government surveillance; to gain real trust, the “digital assistant” will need to be open-source and running on your own hardware, and that’s still a long way to go.

Anyway, it looks like today’s contenders stand no chances at all: Amazon, Google, Yandex; there might be a slight chance for Apple and Siri. Of course, people are strange: as of today, 60% of the web browser market share in Europe belongs to a browser developed by a company with Internet advertising as the main source of profit — go figure…


Comments can be sent as webmentions or by email.