27 Apr 2017
What is: artificial intelligence?
When you hear the term “artificial intelligence”, what pops into your mind?
Probably something like Skynet from the popular Terminator movies, a super-intelligent computer which wreaks havoc on humanity and launches an almost apocalyptic war. Or maybe it’s HAL 9000 from “2001: A Space Odyssey” which slowly turns malevolent over the course of the movie. Or perhaps it’s the mass of robots that attack Will Smith’s character in “I Robot”.
But those are some pretty negative examples. You could also think of the cute and cuddly robot AI “WALL-E” of the Pixar movie with the same name. Or you might think of your beloved Siri – the one friend who never will leave you (unless you lose your phone, that is).
Whatever the case, what’s interesting about modern AI and how it is represented in popular media, is that almost all of these examples (with the obvious exception of Siri) aren’t even remotely close to becoming reality. This might be a shock to you – surely, with the massively powerful supercomputers available today, and all the buzz about big data, predictive analytics, and machine learning, we’re close to AI that is smarter than humans, right?
You could point to Deep Blue beating humans at chess or Jeopardy or Go and say, “Hey, look! AI is smarter than humans! Pretty soon we’ll have robot butlers and the whole deal!!”
Well… yes and no.
The thing is that the key difference between pop culture portrayals of artificial intelligence and the AI of the real world is that in pop culture, AI is represented as a singular program/robot/device capable of performing infinite tasks and solving complex/varied problems as it goes, whereas here in reality AI is much more narrow in scope and much more varied in design.
The artificial intelligence of today (and the algorithms which drive it) can complete tasks and solve problems, in many cases much faster, more efficiently, and more effectively than a human can (take writing email subject lines, for instance). The important distinction here is that those tasks and problems must each be dealt with separately, and each require their own specific algorithm. Rather than one magical AI machine meeting every requirement of mankind, what we actually have is a multitude of machines, each working within its own narrow scope, completing their own specific task, and meeting our requirements as a team.
Confused? You aren’t alone. Let’s delve a little deeper.
What is: artificial intelligence?
Strong AI (or general AI) is what most people think of when they hear the term “artificial intelligence”. Strong AI is a form of intelligence that possesses a human-like cognitive adaptability and ability to deal with uncertainty, along with general problem-solving abilities.
For example, the following situations would only be solvable by a human or a strong AI:
- What direction to take a startup in
- How to deal with a person who is emotionally overwhelmed and needs support
- How to land a plane given unexpected and novel conditions: such as needing to do a water landing
There are three important things to understand about strong AI:
- Strong AI doesn’t exist yet. Strong AI is currently only present in our imaginations and in books, TV shows, movies and other forms of media. Some experts think we’ll arrive at strong AI by 2050, but historically researchers have had a terrible track record of predicting when strong AI will develop. So it’s really anyone’s guess.
- Strong AI is a separate topic and field of research from robotics. There’s no inherent reason that AI of any kind needs to have a humanoid or robot body – this is just a concept that is popular in movies. There are plenty of robots functioning now that use weak AI or human input to control their movements.
- Strong AI isn’t simply a function of computing power. Current supercomputers are much, much more powerful than the human mind at certain tasks – such as crunching, and remembering, large amounts of data, doing fast calculations, or doing repetitive things over and over. However, these computers are also unable to do many things that come quite easily to humans.
As this Punchkick Interactive article “7 common misconceptions about artificial intelligence” states:
“As of 2014, the world’s fastest supercomputer can perform over 11 trillion operations per second. By contrast, an individual neuron can fire and reset itself about 200 times per second.
Yet a human can perform tasks within a second that these supercomputers cannot perform within the same amount of time or are unable to perform at all. This tells us that whatever the brain is doing to perform these tasks, it is going about it in a fundamentally different way than today’s most powerful machines are.”
Weak (aka narrow AI)
Weak or narrow AI performs a specific task very well. Often a task that involves lots of repetition or a large data set. This kind of AI is prevalent today in many different forms.
Siri and Google Now are examples of narrow AI that perform one function very admirably, translating human speech into commands that are then executed on your smartphone. Another example is the software on your car’s ABS that prevents your brakes from locking. It uses data from your car to do this one task very well, but it cannot perform other tasks.
This AI is ‘narrow’ as it’s programmed to do one thing. For example, Deep Blue – IBM’s famed AI system that beat chess champion Gary Kasparov at chess – took years and years of development in order to accomplish this feat. If you asked Deep Blue to play checkers, it wouldn’t be able to play until human engineers programmed that function into it, despite checkers and chess being similar in a lot of ways.
Essentially this type of AI is a computer program that performs a given function that it has been explicitly programmed to do by humans. This kind of AI cannot deal with uncertainties or new situations that have not been programmed into it already.
Narrow AI is seeing more and more commercial use and consumers love its benefits. Our own software, Phrasee is a form of weak AI that uses large sets of data and language processing capabilities to make your email marketing better.
Let’s dive into some other examples of weak AI that are currently on the market.
Amazon’s Alexa: Your personal DJ (and so much more)
You’re lying on the couch, and you don’t know where you put your phone. You feel sluggish, and you want to play some Hip Hop to get your energy levels up.
Luckily, Alexa has your back. “Alexa”, you call out. “Play my Hip Hop Motivation playlist on Spotify please.” The Amazon Echo device (which “hosts” Alexa) hears your voice, interprets the command, and accesses your stored playlist.
Within a few seconds, bass is pumping from your subwoofer. Great job, Alexa!
The Amazon Echo is a device with 7 microphones for hearing your voice that is linked to Alexa, a form of weak AI that processes your commands from spoken word into action.
Here are some of the commands you can ask Alexa to do:
- Alexa, what time is it?
- Alexa, what’s the weather like today?
- Alexa, can you set a timer for 10 minutes? (Useful for cooking and baking as this process is entirely hands-free)
Alexa also has some slightly more advanced capabilities, including:
- Reading your Kindle books out loud
- Ordering you an Uber (and Alexa will even tell you how much Surge pricing is)
- Telling you the most recent scores of your favourite sports team
This could explain why the Amazon Echo and Alexa are so popular that they’re being bought in droves – and even receiving marriage proposals:
Microsoft has its foot in the weak artificial intelligence market as well, with its personal assistant named Cortana. Cortana is included by default on new Windows phones and Windows 10 operating systems, so it’s most likely that you’d be using this AI on your PC.
Cortana is voice activated or usable through keyboard/touch/mouse input as well, and has many similar functions to Alexa, such as:
- Setting reminders for events
- Doing simple calculations
- Giving you directions
- Launching programs
In a head to head comparison involving both Alexa and Cortana, Business Insider contributor Jeff Dunn found that Alexa was the best choice for a smart home and for simple tasks, while remarking that Cortana “is a Bing shortcut.”
***However, both artificial intelligence assistants were able to tell dad jokes:
Cortana: “How many politicians does it take to change a light bulb? Two. One to change it, and another one to change it back again.”
Alexa: “What’s brown and sticky? A stick.”
Do you remember the chatbots from AOL Instant Messenger? You know, the ones that would have lame conversations with you that you could tell were generated by a computer?
Well, AI’s language recognition capabilities have gotten a lot better since the days of dial-up, and many companies are taking advantage of those improvements. For example, chatbots are being used as automated customer service representatives that attempt to answer your questions without involving humans.
For example, let’s say you want to order pizza, but it’s late at night, you’ve been drinking a bit, and you don’t feel like talking on the phone. You can use Pizza Hut’s Facebook or Twitter chat bots to order a hot pie to your house without downloading the Pizza Hut app.
Another example is the chatbot launched by beauty brand Sephora, which asks you some questions about your tastes and then messages you personalised videos and information based off that info.
AI email subject line science
Artificial intelligence functions most effectively when dealing with a relatively simple problem. Could AI write an eloquent, human-sounding novel all on its own? Nope (not yet, anyway), but what about a single line of text?
This was the problem tackled by plucky artificial intelligence startup Phrasee.
Phrasee developed AI algorithms which could produce marketing email subject lines (that single line of text you see in your inbox which describes what an email is about) for the purposes of getting more recipients to open marketing emails. Phrasee’s AI also analysed the performance of each subject line over time to learn which words, phrases, offers, etc. performed most effectively for each of Phrasee’s clients. The algorithm made adjustments accordingly as it went to optimise the language it used.
The AI generated subject lines that outperformed those written by humans in 95% of cases! This increase in performance has resulted in drastic increases in email marketing revenue for Phrasee clients. Such success stories represent a clear indication of the potential for artificial intelligence in the marketing and advertising realm.
What artificial intelligence is (and isn’t)?
So, what have we learned today?
- Artificial intelligence is able to do simple, repetitive tasks much better than humans can – such as crunching large amounts of data quickly.
- Artificial intelligence is able to do very specific functions like interpreting human language (either typed or spoken) and carrying out commands based on those interpretations.
- Artificial intelligence is not able to take over the world, or even solve simple problems that involve uncertainty and/or novel information.
- Artificial intelligence is not the same thing as a robot. Robots and AI are distinct, separate things.