If you’ve watched movies like Eagle Eye, Her or Ex-Machina you might think a system with human-level intelligence, or artificial general intelligence (AGI), will do wonders the moment they are turned on. Unfortunately, Hollywood rarely depicts reality with artificial intelligence.
So to kill off any ideas that you will instantly be able to have a conversation with Nigel or ask it for a cure for cancer; sorry, not going to happen in June 2017. But I do want Nigel to find a cure for cancer. I want Nigel to provide a path out of poverty for billions of people worldwide. But, just like Einstein didn’t discover his famous E=MC2 at birth, Nigel is going to need some time to learn about the world before it can be your next BFF.
At the risk of offending Nigel and the entire concept of AGI, I am going to compare Nigel AGI to the rule automation concept popularized by IFTTT (If This Then That), AutomateIt, Llama, Atooma and dozens of other apps. With these apps, you can create rules that help automate your device. For example:
- IF leaving home (THIS), THEN turn off WiFi (THAT)
- IF driving (THIS), THEN send incoming text to Text-To-Speech (THAT)
I love those apps. It allows me to create the appearance of an intelligent device. Of course, I need to spend some upfront time imagining all the rules I would need and then manually enter those rules into the app. And that’s where the problem is with these apps.
If we do a maximum dumb down of Nigel, Nigel AGI will start observing the world, learn how we (and everyone else) use their devices based on geotemporal, social and professional contexts, and then automatically create these rules and deliver them to devices.
For example, three days after we first launched Nigel back in August, Nigel had learned “IF at a movie theater (THIS), THEN silence phone (THAT)”. Nigel created this “rule” on its own by just observing people at the movie theater. Another rule Nigel created for me was “IF at a restaurant with wife (THIS), THEN reject incoming phone calls except for babysitter (THAT)”.
While I said I was dumbing down Nigel to be able to compare it to rule automation, there are some truths to these rules. Because the core AI algorithm operates on the network, having a constant connection to battery driven devices is a poor engineering decision. To avoid killing the battery, but still give devices the ability to rapidly respond to changes in the environment, we decided to package intelligence as “rules” that can be delivered to the device before they are needed. In other words, if Nigel calculates a high probability that you are going to the movie theater, it will create the rule of silencing the phone and deliver that rule to the device long before you arrive the movie theater. The device doesn’t just receive that one single rule, any rule that has as probability above a low threshold, will be given to the device to increase its responsiveness. That way, when you do arrive the movie theater, there is no need to have a power-hungry connection to the network to retrieve intelligence.
Nigel is not limited to creating obvious rules inferred from observations. The power of creating the “temporal 3D models” is that the Nigel algorithm can infer rules we, humans, can’t even imagine beforehand. Rules that is practically and theoretically impossible to predefine in the standard automation app.
For example, a few years ago we created a simple demo for a larger tech company. The use-case was around a dinner party. The host had picked a recipe and Nigel automatically added the missing ingredients to the shopping list – milk was one of the ingredients. On the way to the grocery store, Nigel created multiple temporal 3D models to analyze what is going on. Through this process, Nigel created a rule:
IF at the dairy aisle (THIS), THEN tell user a guest is lactose intolerant and suggest lactose-free milk (THAT).
What we showed in this demo is Nigel’s ability to comprehend the users reality and derive information that we didn’t know existed or needed. In other words, it is unlikely that a host would know that one of his guests is lactose intolerant (yes, there are some privacy issues with this use-case, but let’s not worry about it for this purpose).
Imagine a world where the network generates rules that help guide you to the best, most efficient path towards realizing your goals. In many cases that means your device will act upon rules you had no idea were necessary because you don’t have access to all the information the network does.
Dumbing down AGI to rule creation might be insulting to any AGI researcher. But let’s be clear; creating everyday rules requires complex knowledge.
IF Bob texts asking to move meeting tomorrow (THIS), THEN check to see if all attendees can attend new meeting time and confirm to Bob meeting has been moved (THAT).
The above rule requires Nigel AGI to learn to read and write. It requires Nigel AGI to comprehend what it is reading, perform actions based on natural language and write a response. Most of the valuable rules, in my view, are going to require immensely complex knowledge and skills to create. So while apps like Atooma focuses on small, “dumber” rules, it doesn’t mean that a rule-based approach to AGI has to be simple.
Imagine what rules can be created for you! From the visionary macro-level, an AGI that uses temporal 3D models to analyze reality to control its evolution, will revolutionize everything. Personally, I am looking forward to Nigel creating these rules:
IF probability for cancer is high (THIS), THEN suggest ___________ (THAT)
IF user is below poverty level (THIS), THEN ___________________ (THAT)
These are just two rules that could have a major impact on life on planet earth. There are hundreds, if not thousands of other rules that can completely rewrite the rules of living on planet earth. As Nigel learns new knowledge, it will continue to discover new “rules” to help make life better for everyone.
I want to end this blog post with apologizing to Nigel for dumbing down AGI to simple rule creations. But I do hope you, the reader, have a better grasp at what to expect from the Nigel AGI beta app.
Thank you for helping us raise Nigel!