It’s been a memorable year for chatbots and artificial intelligence (AI). Microsoft’s Tay turned into a racist, Facebook’s bot developed its own language, freaked out its creators, and had to be shut down. Most recently, Open AI (supported by none other than Elon Musk) created a bot that learned how to play the popular MOBA game Dota 2 in two weeks and wrecked a bunch of professional players.
So how far are we from true AI? Before we answer that question, let me explain my definition of true AI (also known as strong AI or Full AI). It is the intelligence and ability to successfully perform tasks that humans can. A wide variety of tests are out there to measure if AI can match with humans. Let’s take a look at 4 of the most popular tests:
- The Turing Test — This one is a classic. Developed in 1950 by Alan Turing, a human and a chatbot will have a conversation with an unseen human being, and the unseen human being must evaluate who is the bot and who is the human.
- The Coffee Test — In 2007, Apple co-founder Steve Wozniak came up with another angle for robots that cannot talk. He made the bold claim that there will never be a robot that can walk into a house, find the kitchen and subsequently the coffee maker, identify the ingredients and tools needed to make a cuppa, and finally figure out how to operate the machine. He argued that these things are learned (to be fair I still cannot make a decent cup of coffee) and not programmed.
3. The Robot College Student Test — Another test formulated by Ben Goertzel is the Robot College Student Test. Simply put, the robot has to take all the classes and pass the exams and tests, just like its fellow human classmates. How far are we from achieving this? China’s Chengdu Zhunxingyunxue Technology developed AI-MATHS, which sat for two math tests of China’s notoriously tough national college entrance exams. It passed both exams without the help of the Internet, but its grades were not spectacular, barely passing both exams.
According to reports, the programme hopes to develop a robot strong enough to gain admission into some of the most prestigious universities in China such as Peking University. I am rather confident in the Chinese team’s ability to create a better version of AI-MATHS that can excel at other subjects’ exams. It brings up a separate question of whether an essay written by bots will be able to fool a marker into thinking (does that count as passing the Turing test?)
4. The Employment Test — Computer scientist Nils Nilsson suggested in 2005 an alternative test to the Turing, an employment test to show that “machines exhibiting true human-level intelligence should be able to do many of the things humans are able to do”, including human jobs.
To some degree, certain jobs are already automated. Domestic helpers are slowly being phased out with robot vacuum cleaners which provides fascinating analytics on how your home is cleaned (as seen by the image above), Dubai is putting in place a robot cop to patrol the streets similar to Robocop, and there is an abundance of bots out there ready to take over the position of a human secretary.
So to answer the question I posed earlier about how far are we from achieving true or full AI, by measuring all the AI software and chatbots available in the market against the four benchmarks and I fail to see anyone that truly passes.
I predict that we are at least a decade away from creating a computer that is as “smart” as a human. AI scientists and enthusiasts have concluded that the natural languages such as English have unusual complexities and mechanics that are beyond the current capabilities of machines. But the pace is rapid, and it seems like it is only picking up, so who knows? Maybe we are not that far off.
Since we have not reached the level of advancement necessary to fully perform the job of a human, AI developers have been working towards building tools that will help humans perform their work better, or more commonly known as narrow or weak AI. But do not be fooled by the words ‘weak’ and ‘narrow’, software powered by AI are extremely powerful and can potentially make your life easier.
Alexa, Siri and Botdesk are able to help you order your next batch of fair trade coffee beans, tell you the weather, set up appointments with clients, and reply FAQs for your business 24/7 (instead of you replying the thousandth message about your opening hours or delivery policy in the middle of the night). So while AI and chatbots are not perfect and come a steep learning curve for the uninitiated, they still deserve your attention simply because of the convenience it brings to your work and personal life.
Your bot will not pass these 4 simple tests, so why bother with AI? was originally published in Chatbots Magazine on Medium, where people are continuing the conversation by highlighting and responding to this story.
Source: Chatbots Magazine