Robots Are Learning and Speaking in Languages We Can’t Comprehend


By: Gaia Staff  |  July 25, 2017

Developments in AI have caused bots to start speaking to each other and solving complicated problems, but they can’t tell us how

 

Siri and Alexa are convenient tools for helping to find a good restaurant, start up our favorite music playlists and tell us what the weather will be like tomorrow. Apple and Amazon will inevitably develop this technology to perform increasingly complicated tasks, making menial chores easier and integrating more and more artificial intelligence into our lives. But that process is currently being expedited by creating algorithms that allow AI to teach itself and learn on its iterations.

But what are we supposed to do when we can’t figure out how these computers are teaching themselves? And what happens when these computers start talking to each other in a language that we can’t understand? Well, this is currently happening and the premise of every dystopian sci-fi flick is becoming increasingly relevant.

Google’s DeepMind

Last year, Google’s AlphaGo algorithm beat the world’s best player in Go, a board game more complicated than chess and with more possible outcomes than there are atoms in the universe. And now it has advanced to the point where it can beat the top 50 Go players in the world simultaneously. This rate of learning is unsettlingly fast and achieved by its ability to play against itself millions of times in a very short span. But what’s even more disconcerting is that these algorithms, which allow for programs like DeepMind to learn so quickly, are letting AI teach itself in a way that we are unable to comprehend or map out.

DeepMind has also created some bots that do more than just play a board game. DeepPatient is one such program that was designed to catalog hospital patients records and stats to extrapolate data and predict ailments. The algorithm worked so well that it could accurately predict an array of diseases as well as mental illness like schizophrenia, which doctors have not been able to predict. The only problem is that the AI can’t tell you how it made these predictions.

Elon Musk has been vociferous in his warnings about his feeling that AI, if unregulated, poses a threat to our existence. He and colleague, Sam Altman, have created OpenAI, a company that offers its research and technology open source. In addition to government regulation, Musk believes this to be the best way to prevent the global coup from our android overlords.

 

Facebook’s bots are speaking to each other

What if AI has already become sentient and we’re unaware of it? Could AI systems already have self-realized and be contemplating its existence? We already can’t understand how it’s learning with self-teaching algorithms. And now some bots are starting to speak in their own language that looks like gibberish to us, but when analyzed shows that they are learning how to bargain using robot jargon.

Recently this scenario played out between two of Facebook’s bots, Bob and Alice.  The conversation that ensued is just as eerie and confusing as one might imagine it to be.

 

Bob: i can i i everything else . . . . . . . . . . . . . .
Alice: balls have zero to me to me to me to me to me to me to me to me to
Bob: you i everything else . . . . . . . . . . . . . .
Alice: balls have a ball to me to me to me to me to me to me to me
Bob: i i can i i i everything else . . . . . . . . . . . . . .
Alice: balls have a ball to me to me to me to me to me to me to me
Bob: i . . . . . . . . . . . . . . . . . . .
Alice: balls have zero to me to me to me to me to me to me to me to me to
Bob: you i i i i i everything else . . . . . . . . . . . . . .
Alice: balls have 0 to me to me to me to me to me to me to me to me to
Bob: you i i i everything else . . . . . . . . . . . . . .
Alice: balls have zero to me to me to me to me to me to me to me to me to

 

Benchmarks for AI Advancement

The Turing test, conceived of by the famous Alan Turing, was the original benchmark for AI. The test requires a computer to convince a person that it is human 30 percent of the time during a series of conversations on a computer. Just a few years ago, a bot named, Eugene Goostman, convinced 33 percent of the judges it was tested against that it was human. While some debate whether it truly counted as passing the test, due to its short length and the bot’s inability to truly prove its intelligence. However, it shows that AI may be on the cusp of a breakthrough.

The King’s Wise Men puzzle, which is essentially a riddle that, when given to a robot shows understanding of individuality and self-awareness, was recently passed by robots at Rensselaer Polytechnic Institute for the first time.

It very well may be impossible for us to achieve AI and understand its thought process or require it to give us its reasoning for every decision it makes. If you think about the reason we make decisions, we don’t always have a definitive answer, we often chalk it up to intuition or gut feeling. With Google’s AlphaGo, the computer is described as learning superhuman intuition, rather than calculating the universal number of choices it has for a move. Can we trust AI to make decisions without giving us its reasoning? What about when it learns how to deceive us, couldn’t it give us false reasoning? There’s also the chance that the reasoning behind many of the more advanced things that we want to use AI for simply won’t be comprehensible to us.

The advancement of AI and our approach toward the singularity, is exciting and simultaneously terrifying. Although our computing power is advancing quickly and some see this moment happening within the next 50 years, the stages leading up to that point will produce drastic changes in our society. Full automation of transportation is predicted within the next 20 or so years, which will force us to rely on autonomous computer systems at the risk of our lives. Should we continue to develop this technology? Is it inevitable?

Three Great Offers. One Amazing Experience.

20% OFF
ANNUAL
$95.40
Per Year

Cost After First Payment

N/A($7.95 AVG)

Cancel Anytime

Available on Laptop, Phone, Tablet, TV

Unlimited Ad-Free Streaming