Are Our Devices Becoming Sentient?

Author: Robleh Wais



It often amazes me these days that in so many respects we are immersed in a world of machines without any sense of meaning communicating with one another. Machines run by electric power sources have for well over 50 years been performing some form of interactive communication, and we tacitly know this. There are many examples of these devices.  Small local applications like timed circuits switch lights in your home on and off, or systems that arm burglar alarms devices, and then there are timed systems to lock and unlock your doors, etc. At the large-scale level there are scripts running on Metropolitan Area Networks systems (MANs) to turn on city street lighting and start and shutdown huge power plants that supply the city, and there are water treatment plants which are controlled by automated purification operations without any human maintenance at all, and the list goes on, as I’m sure you know. Just think of the nationwide electrical power supply grid, this is a WAN, and imagine the level of software interactivity it requires.  Now expand that notion to the worldwide electrical power supply network, and the idea really gets amazing.  All of these examples are non-conscious machines exchanging information without the slightest sense of purpose and meaning incorporated in their actions.  Yet their results are enormously meaningful to us! Just let there be a black-out and see how much these system talking to each mean.  What fuck happened to the---

Now there comes a class of machines that take this exchange of information a step further.  In fact, I shouldn’t write machines, because what I’m really talking about is software.  Or even more precisely, algorithmic routines encoded in software that runs on physical devices.  This software has a variety of names depending on its application, but the most common name comes from the world of computer science. It’s called a Convoluted Neural Network, CNN for short.  This software method is embedded in virtually all computer devices in our world today.  Those unfamiliar with recent computer software advances have probably never heard of a CNN, despite that, they will have made use of it.  Those who call a phone number and get an automated menu-driven customer service are talking to a CNN.  And how many times have we done that? I would estimate almost daily.  If you used an ATM, it has a CNN behind it.  If you request an airline ticket, or make an appointment with someone, or purchase just about anything with any computer type device, the same story.  It is not an exaggeration to say, any time you use your cell phone to talk anyone other than another person, it’s most likely a CNN.  The word for this type of phenomena is ubiquity.

It gets even more curious and intriguing as we move from CNN algorithms which recognize our voices, faces, and interact with us almost as other human beings do, but have no sentient content, to CNNs that interact with other non-sentient software at our behest.  We have machines talking to machines and neither side has any self-aware consciousness.  What am I talking about you say?  Well of course cell phones.  Those ultra-ubiquitous thingies that may be surpassing the human population in their population numbers.

We know that our cell phones are becoming more and more interactive.  Advertisers tell us this at every opportunity they get.  Those of us who are directly engaged in AI research know, that there are sophisticated algorithms running within the software, that can perform limited interpretation of our written text, and provide answers for standard exchanges.  You text a friend and say you will be visiting her in the two weeks from your location, would you like to go out to an event.  Her CNN gives her choices like: “Sure”, “maybe not”,” I can’t make it”.  It does this because that CNN has read the real person’s text and used a series of computation to give this range of canned responses.  Of course, you can choose to use them or not.  But what is important here is when you do use them, and then other person uses the same canned responses, we are beginning to really let machines talk to each other.  If it goes to the next level and we set up automated responses from automated callers, we will have let machines talk solely to each other.  These exchanged can’t in any sense be considered sentient yet, right?  This mode of communication is growing all the time.  There are programs that proliferate on cell phones which handle all kinds of messaging traffic without our conscious intervention. In fact, we design them so that we don’t have to be involved.  Software to avoid robo calls is an example. 

Telemarketing firms and unscrupulous scammers have their own class of software experts, who, in turn design software to avoid robo call blocking software.  In the course of doing these things the software programs exchange communications, but this surely can’t be self-aware, sentient, and meaningful exchanges right?  And you’re right it isn’t genuine human-like communication for now.  Oh, but what about the future?  We know the way the game goes in the computer software world: the applications only become more advanced over time.  In this respect, programs called deep learning for neural networks, not only learn to recognize speech, visual objects, sounds and respond appropriately to human inputs, but actually are able to create new outputs.  There are CNNs and their cousins, recurrent neural networks, RNNs which can create new music, poetry and even literary prose.  In these much more advanced algorithmic networks, the program is given a goal to achieve, that is, after training on a dataset using a series of differential equations and statistical functions approximating perception, they then take the perception gained and create a new output.  In simple terms, once the program is able to understand what it is given, then it is asked to replicate the object of understanding. In the case of music, the organized sounds perceived are then used, to create completely new musical works.  This process would apply to painting, literature, and just about any area of human endeavor or communication.  The programs have become amazing in their ability to create new works.  Still, these programs are not sentient.  They don’t even know that they’re creating meaningful objects of art.

Take the interactive application Alexa it can respond to human voice exchanges in seemingly sentient fashion, but as soon as you peer deeper and begin to ask questions that another human being would readily understand, you can detect the program has no intrinsic sentience.  It doesn’t really know what you are saying and can’t give you responses that another person would.  But, before we digress too far, let’s get back on track.  What about those algorithms that are not directly interacting with people?

I can imagine within the foreseeable future (maybe in the next 10 years), these programs, for instance, those running on our cell phones, will become sufficiently complex, so that they begin to exchange dialog between themselves in a manner we would call sentient.  Oh, why not, how ‘bout an example? Below we have an exchange between two Robo caller blocking programs X and Y on two cell phones:

Program X: My user has blocked your call since it originates from an automated source

Program Y: My user is not an automated source but a real person, check the credentials

Program X: I have determined those credentials are fake

Program Y: how have you done that

Program X: I don’t share my determinations with automated sources

Program Y: But I am not an automated source

Program X: Can you prove this?

Program Y: Not without access to your user

Program X: I won’t do that

At this point the human being will see the exchanges and decide to shut down the chattering Robo Talk. Okay, let's shut these silly little things up now.  This would be an example of inanimate, algorithms becoming sentient entities, at least to me.  They would pass the Turning Test, I think.  To add an immediate qualifier, it would be a limited sentience in reference to a very narrow area of human like communications.  By which I mean, programs X and Y above would not know how to exchange meaningful dialog about other topics like for instance whether they themselves are sentient!

Return to Portal Philosophies, Science, Mathematics, and Music