Artificial Intelligence is watching you

Original author: Mike Szczys
  • Transfer
image

My phone has learned to understand me, but it still does not have enough intelligence to understand what I want. We already have the hardware and software capabilities for this - there are not only social opportunities.

Personal automation is still underdeveloped. I have Google Now on my phone. Every month, Google Now reminds me to pay bills I’ve already paid. He does not see that I have paid them, he just sees the letter I received and the due date. I act out of habit, and pay bills on the last day of each month, even if there are still a few weeks left before the due date. And this is the easiest thing a computer could learn. But this is a system without feedback, which means no training.

I recently read a couple of greatarticles on the current state of development of AI, shed light on this problem. According to them, the current level of personal automation can be described as “weak AI”. And I need a generalized AI (OII) for personal needs. But such an AI cannot be expected, and here is why.

Blind AI


Like most people, my phone has become a part of me. Although I use a desktop computer for many hours a day (with a separate monitor and keyboard), quite a large part of my life passes through a 5.2-inch touch screen. Google constantly watches me, but so far it is dedicated only to a small part . he sees the monthly notices of payments mentioned by me earlier as I use gmail, but it does not follow my browser sufficiently closely to know that I have to pay for them.

In life, everything that happens needs a boost. If my payments on accounts no Obra Noah connection, it is not surprising that the older I get notifications about them. This means that the current level of automation does not help, just annoying. It can not keep track of everything that I do.

How to follow a person so as not to scare him


image
Kayla constantly listens.

In many cultures it is not customary to gaze at people. That is, that, it is possible or it is impossible to look at the person, depends on the moment; there is a maximum allowable period of time during which a person can be considered; and the rules governing this game change as the game progresses.

However, almost all people are able to learn how to play it. Even completely unfamiliar people are able to recognize when you need help, and whether they need to offer help to you. This is the cornerstone for the development of personal AI. And at the same time, this is an incredibly difficult task.

The easiest way to follow absolutely everything that the user does. Through this approach, we get much more data, but this is a frightening practice, causing tons of ethical questions. No one has ever been constantly monitored - for such a process there is no interpersonal paradigm. Existing technological paradigms do not cope with this task. Just last month, the German authorities recommended that buyers buy a doll called “Cayla” to destroy the microphones in it . Her microphone is always listening, processing the received information through the speech recognition system, despite the fact that the service itself is located outside the country.

Whatever the scarecrow is, privacy is a major problem on the way to let the system keep track of everything you do. Leakage of such information will be a gold mine for "identity theft." Will your AI understand that it should shut down every time you go to the toilet, hospital, other sensitive environment? Can you trust him that he disconnects at such moments?

My whimsical imagination already draws a scenario in which your AI becomes so clever that it starts to blackmail you (a thought similar to the ideas of Douglas Adams). What is more likely when your personal assistant gets to know you well enough and proves that he can help you do your job more effectively, he will be promoted to a personal manager. Will you still be an effective team in this case?

Machine learning as a social norm


image
Sethling's Neural Network is learning to play Super Mario World

Machine learning is the key to amazing accomplishments. But if you understand a little how it works, the problem becomes immediately visible. The machine can learn to play games very well, but it needs to see all aspects of the game and give specific parameters for success, such as points or collected items.

To make personal AI really useful, it will require almost unlimited access to data collection through observation of your daily life. But this is not the end. The AI ​​cannot repeat your Monday’s routine over and over as it is possible with the level of a computer game. In this case, for successful machine learning, the algorithm needs to exchange data on a large number of people in order to obtain a useful set of parameters. Technically it would have worked, but it would be a network of epic size peeking over people. This is no longer an ominous valley, but simply the plot of a horror movie.

We have already encountered problems when collecting so much data. Social networks are machine learning without the benefits of AI. Millions of people publish, from their point of view, harmless information about themselves on countless platforms. But big data processing turns this harmless information into predictions about the behavior of population segments.

If people experience dopamine delight in sharing this data, how will effective personal AI affect them? He will simultaneously be your friend, adviser, confidant. I take off my hat to Charles Stross, who described a very terrible version of AI in the novel Accelerando. In the book, AI takes the form of a realistic robot cat. It is very easy to underestimate abstract intelligence.

Even with access, the AI ​​cannot see


image
The Google project “Inceptionism” gives out drug-related images, but it still doesn’t understand what it is looking at.

Current technologies may allow you to create a unified data set, watching everything that happens on your different computers and mobile devices. AI can record the audio of your life. And even record a video.

But even if you have full digital access to your life, it will be very difficult to make sense of what you are doing. This is not an insurmountable obstacle, but with current technologies such a task will definitely require remote processing of data at high power. And such attempts are already being made by systems like Amazon's Echo, Apple's Siri, Google's Allo, and children's toys, such as the previously mentioned Cayla and Barbie from Mattel.

There is no video recognition yet. This is the cutting edge of technology (the development of robots and military robots), and this opportunity will soon appear. As in the case of voice recognition, services like Google Cloud Vision have already appeared , depending on the system of restrictions: the orientation of the object in relation to the camera, the level of illumination, the presence in the set of objects for comparison, etc. But I do not think that in the foreseeable future, a reliable computer vision will be a suitable source of data for personal AI.

And for our life is a problem. How does your AI know who you are talking to? Without seeing what you see, it is very difficult to collect data by context. The most obvious way to implement the entry of such data would be wearing cameras like Google Glass. How this experiment ended, we all know. Maybe Snapchat attempts will fix this situation.

What could we have but not


Well, I have already written enough complaints about the problem. If you manage to solve all these problems, what do I really want? In short, I want my intelligence to complement.

If I talk in passing with my wife in passing, discussing a musical preparing to leave, I want my personal AI to remember this and report on the moment of the start of sales of tickets to it. Besides, I want him to know where I like to sit, how much I am willing to pay, check my calendar and my wife’s calendar to choose the perfect day, and just ask me for permission to buy. I want him to know that we usually combine the trip to the show with dinner or with the use of drinks, so that he would study the history of our trips to restaurants and figure out where we would most like to sit this time.

But I also want to keep my private life secret. I want to be human and live my life. So I’ll tear away from these dreams of a brave new world and enjoy what I have: access to information, which could not have been seriously dreamed of 30 years ago. Technologies will continue their development, and we will get our advantages from this. But so far, these technologies cannot monitor us fully enough to make AIS a part of your daily life. But people will try to do it, and it will always be interesting to read about such attempts on GeekTimes.

Also popular now: