Intern’s Digest: Artificial Intelligence
In the more romantic frames of cinema, Artificial Intelligence (AI) is often depicted with a living, talking, sentient body. Be it a computer with emotional intelligence working to its own agenda like HAL, or Oscar Issac’s robotic human companions so lifelike they are possible to fall in love with. It’s a buzzword in all tech articles and marketing press at the moment and the spectrum on which innovations are deemed to be considered AI is becoming broader and broader. So what exactly is considered AI? Where is the technology going? And what are the mainstream implications of it?
What is AI?
AI in its broadest sense is an algorithm or machine or robot that can make its own decisions, within the confines of its predefined rules. It can play chess, but only chess; it can make a car, but only that car; it can find a needle in a haystack, but only the needle you tell it to find, and in only that haystack. It can be taught to learn, but will be told from which sources to learn from; it can generate its own thoughts and outputs (tweets, articles, social media posts, movie scripts, etc), but will be given a template to build its responses from; and can attempt to think for itself, but will be told what to think about and from which to learn how to think on that subject. The current state of AI involves large ‘think boxes’ that will consume all possible information on a predefined subject, to the point where it will be able to produce outputs autonomously.Alan Turing wrote a paper on the notion of machines being able to simulate human beings, and a great wealth of academic time has been spent debating whether it is possible for a computer to ‘think’. The above examples are designed to demonstrate both the usefulness and limitations of AI. A typical example of exploring a machine’s ability to ‘think’ is the ‘Chinese Room’ argument.
“Imagine someone is locked in a room, where they were passed notes in Chinese. Using an entire library of rules and look-up tables they would be able to produce valid responses in Chinese, but would they really ‘understand’ the language? The argument is that since computers would always be applying rote fact lookup they could never ‘understand’ a subject.”
A similar argument uses the Turing Test – a benchmark for artificial intelligence to be considered genuinely ‘intelligent’. In Turing’s 1950 paper “Computing Machinery and Intelligence” he puts forward the idea of a machine being so alike to human intelligence, it would be indistinguishable by a suspicious judge. In his original paper, he referred to the test as the “Imitation Game”, and its inception is credited with spearheading research in maths, physics, philosophy and engineering to bring us to the point today of machines “passing” the Turing Test. The test itself has taken on many different forms as AI has become more complex but the original test involves an interrogator who must distinguish between a human and a machine by asking questions. The responses are typed and seen blind.
To push the field further there is now an annual competition for the best AI algorithms with a $2000 for the best entry. If you do actually manage to pass the Turing Test then there is a $100,000 prize. The exchanges that some machines have made with judges are remarkably similar to human interaction. The 2005 winner is a permanent website and I highly encourage playing with it.
The Current State of AI
The reality is however that the AI that we see entering mainstream practical use will not be a Turing Test beater – but just like NASA pushing boundaries to land on the moon the fallout from the technological advances will trickle through to the masses. There is no doubt that the ability for computers to seek out information for you, compute large quantities of data, or automate complex business processes is a huge technological advancement. What we are seeing are the very first stages of AI, which is involving A LOT of trial and error.
Microsoft’s automated twitter account “Tay” took just 48 hours to start tweeting racist, misogynistic and derogatory drivel (the algorithm taught itself to learn how to tweet by learning from the rest of Twitter, so this is not entirely surprising – although calls of foul play were raised). Facebook recently removed any human touch in its ‘trending’ algorithm. Stories that are relevant to a user and being shared/liked a lot are pushed to the top of the trending sidebar. The teething problems weren’t received well in the press, however; promoting false news stories about 9/11 and a fake story about Fox News anchor Megyn Kelly.
This isn’t to say the technology isn’t useful, or that it is inferior to its human counterpart – it is just the early stages, and the full capabilities are only just starting to be explored. Scriptbook, a US startup, is using the technology to sift through the thousands of scripts they receive and pick out winners to turn into movies. The Republican National Committee are using an AI bot to sift through the hundreds of hours of footage, press articles, and photos of Clinton to find her most weird, distorted and unflattering moments. The GOP staff are then cherry picking their preferences and sharing them on social media. While not everyone’s favourite use of the technology, it does demonstrate the abilities in audience and social media monitoring and how real-time targeting can be put into practice.
The Future of AI
The use of AI by the GOP is just one example of hundreds being used right now to take advantage of the wealth of information available to marketers and advertisers. We know that programmatic media is providing us with a vast quantity of information on how people interact with websites, social media, and advertising. What we haven’t fully incorporated is a method of breaking this down into something we can comprehend and extract insight. This gap is where AI can come into its own and is a topic about which the industry is clearly excited.
In an article for WARC Mark Holden, the Worldwide Strategy and Planning Director atPHDsaid: “We will be witnessing media strategies, creativity and CRM initiatives being created before our very eyes: strategies and ads will be an epiphenomenon of the system – as in, they will emerge from it.” In this respect, AI lends itself to the data and numbers that media agencies have at their disposal. We are already seeing programmatic media buying taking over traditional methods to hyper-target their audiences.
There are possibilities for creative deployment of the technology too. A few weeks ago Condé Nast announced that they would be using IBM’s Watson supercomputer to “help build and strategise social influencer campaigns for brands”. In Lehman’s terms, they will use the computing power to find closer matches between social media celebrities and brands through monitoring interaction, brand identities (or socially perceived identities), and a matching tone of voice.
In 1000 words it’s hard to grasp AI and its full capabilities, so this is very much just a surface overview. What we have found is that there are vast possibilities. The AI mentioned here is just an environment in which a greater ecosystem lives; we haven’t talked about Augmented/Virtual Reality, realistic AI robots, or the other hundreds of offshoots possible. The Drum recently hypothesised about the future of AI: “why wouldn’t it be able to analyse every ad your agency or brand has ever run, and then evaluate your latest campaign ideas against it? There is potential for algorithms that could say things like, “this campaign was successful for factors such as strong copy, eye-catching visuals, or eliciting strong emotional reactions on social media”.”
The topic of AI is dense, growing and difficult to get one’s head around. Exploring the topic is however incredibly fun and exciting, and I highly recommend anyone interested in doing so (this weekend is New Scientist Live at the ExCel centre if you want more right this second!).