The term Artificial Intelligence (AI) was coined in 1956 when American computer scientist John McCarthy defined it as “the science and engineering of making intelligent machines, especially intelligent computer programs”. AI, also known as machine learning, now refers to a machine’s use of algorithms to understand human activity and provide a response to it. There are few industries that are exempt from the influence of AI and advertising and marketing are at the top of the most-affected list. Most recently, BrainScape, PHD’s annual thought leadership conference, focused on AI. During his talk at the event on “AI: Outside of The Black Box”, Kevin Slavin, assistant professor and founder of the Playful Systems group at the Massachusetts Institute of Technology (MIT) explained, “Any industry that has to make decisions – which is every industry – is interested in Artificial Intelligence.” For advertising, AI is becoming more inserted into the consumer journey as technology advances, making personalization the gateway to conversion. Thus, the time for agencies and brands to first understand it and then utilize it is now. Some believe that the development of AI will result in a machine takeover; however, this is not likely to be the most immediate outcome. A mix where machines get better at performing human functions so that humans can focus on the aspects of their jobs that machines cannot do seems like the most probable and effective reality.
Birth of AI
The concept of AI has actually been around since 1943 when mathematician Walter Pitts and neurophysiologist Warren McCulloch wrote a paper describing an artificial neural network and how it could be used to calculate any computable function, even suggesting that such networks could be self-learning. In 1951, Marvin Minsky and Dean Edmonds actually built a neural network computer, which could simulate only 40 neurons – versus billions in the human brain – but it was a start.
It only came to be known as AI in 1956 when McCarthy defined it as such and mathematician Alan Turing paved the way for researchers to join forces with the goal of building an electronic brain. To put things into perspective, AI was first defined in 1956 – the same year the first computer hard disk was used. Just one year after the invention of the microwave oven and TV remote control, the world – or at least the minds leading it – was already prepared to move into the field of AI.
Officially, AI research came into existence at a Dartmouth College conference that same year, which was attended by the likes of McCarthy, Marvin Minsky, Allen Newell, Arthur Samuel and Herbert Simon, who became the pioneers of AI research.
Over the years, AI has received its fair share of criticism, even provoking governments to cut investments, which resulted in intermittent periods of little to no funding – now known as the “AI Winter” – between the 1970s and 1990s.
In PHD’s latest book, Sentience: The Coming AI Revolution and the Implications for Marketing, Kevin Kelly, Wired magazine’s founding editor, says there are three main reasons why AI is moving away from a topic of academic interest or a tool used by advertisers and manufacturers, and becoming an everyday part of consumer lives:
1. Computers are becoming cheaper and more powerful. According to Moore’s law, computers become twice as fast and half as expensive every 12 to 18 months.
2. Social media and mobile devices are resulting in the availability of more data that can be used to train computers.
3. Algorithms have been improving ever since the introduction of deep learning* algorithms in 2006.
*Deep learning refers to a machine’s ability to pick up on patterns and learn autonomously.
The entire history of AI, through all the ups and downs, seems to have been leading to the precipice on which a portion of the world seems to be standing and peering over. It appears that any company with a stake in the present and an eye toward the future should be gearing up for the world of man plus machine.
Whether or not people are aware of it, AI is already present in humans’ every day lives. In some cases – certainly not all – it is not even that difficult to find, if you know what to look for. Each of the talks at this year’s BrainScape event had one specific thing in common: current examples of AI.
During his talk at the event, Slavin discussed the importance of machine-learning algorithms to the shape of modern AI. PHD’s Sentience indicates that these algorithms are quite unique because, “rather than telling the computer what to look for, you give it some examples and it finds the patterns on its own” (Sentience, page 25). As one example of this, Slavin pointed out that 60 percent of what users watch on Netflix is generated through algorithms and he noted that “algorithms determine what we care about more than we are aware”. Another accessible case he provided was that of Forbes Online, which uses AI to write articles that are based purely on facts. He explained that any Forbes article, which does not have a byline, is machine-generated. The last talk of BrainScape 2015 was driven mainly by a real-time presentation of how Google’s Android operating service and search technologies aim to use AI to make the lives of its users easier. Google Zurich’s principal engineer, Behshad Behzadi, took the audience through some AI processes on his mobile phone that are not only learning to pick up content in voice commands, but have context functions. When he asked his phone, “How tall is this building?”, the AI voice responded with the precise height of the Burj Khalifa, the building in which he was giving this talk.
One of the next developments of today’s AI will be the advancement of highly individual Virtual Personal Assistants (VPA). VPAs could drastically change the way that advertisers go about their jobs, as ads may no longer be placed on a website or in a magazine, but could go directly to the individual user to whom the product is personally applicable.
In fact, some advertisers are already using AI to inform new campaigns. In July 2015, M&C Saatchi, Clear Channel and Posterscope partnered to create their interpretation of “the world’s first-ever artificially intelligent poster campaign” for Bahio coffee in London. The poster can assess the audience in its vicinity and adapts ads based on their reactions. An ad that is successful moves on to the next level and one that is not is deleted. The data collected on the people is not stored.
Before understanding how AI has come to play a role in advertising and tech companies, it’s important to define the “intelligence” in Artificial Intelligence. The most popular – and probably the most criticized – way of doing so is the Turing Test introduced by Alan Turing in his 1950 paper Computing Machinery and Intelligence. Instead of questioning if a machine can think, the test questions if a machine can deceive a person into believing it is not a machine – that it is, in fact, a human being.
Fast-forward to today and the semantics of “intelligence” are still being contested and broken down into more specific categories, like narrow and general AI. But, for the purpose of discussing how it can be used in the advertising industry, the most basic definition is a machine that can do things that we normally associate with human intelligence. There are holes even in this explanation and so, it is further defined in Sentience as follows: “Intelligence has many definitions, but most of them include a few basic ideas: the ability to learn from experience, the ability to apply that learning to new problems, the ability to apply logic and the ability to think abstractly” (Sentience, page 15).
Currently, AI in advertising is being used mostly for ad targeting, which, as Klint Finley, writer for Wired, notes, “is nothing new at this point; that’s been going on for a decade or more now.” It’s also being used in analytics, which he humorously describes as a “fancy word for saying research”. JC Oliver, Microsoft’s global head of innovation, explains how AI is immediately applicable to advertising, saying, “We have the ability to understand what people are doing and to serve up content or messages based on their behavior. This is why programmatic has taken off in a big way.”
In short, AI isn’t being adopted throughout the advertising industry in a huge way just yet, but experts have high hopes for its eventual capabilities. Soon, one function of AI will be to give consumers access to everyday products: a VPA will remind its user when they are out of milk or toothpaste. However, the limitations expected for AI’s potential in the consumer market will have to do with desire. AI will be able to help people get to what they want and have previously purchased, but what about items people don’t know they want?
Based on data collected about personal preferences, AI experts believe that telepathy is one aspect of AI that will get better and better. Explaining the sentient road down which AI will soon be traveling, Haroon Syed, director of digital data solutions at Annalect, paints an interesting picture: taking the desire for a drink as an example, he says, “If sentience came into play – with an infinite number of sources like the temperature, the humidity, how often you perspire, how much you drink in a day, how many calories you burn, the amount of energy you consume, etc. – it could start working out correlations, put these numbers together in fractions of seconds and, then, it can know that you may desire this.” However, he does admit that this is as far-fetched as it is far away.
Syed also highlights the difference between need and desire and how the traditional creative process of advertising will maintain its upper hand for the foreseeable future: “The need will become technological; the desire will still be maintained as artistic methodology.” AI will be based more on rational decision making, Finley points out, but not all purchase decisions are made rationally. He says, “These two things are at odds with each other: AI is mostly predicated on context and behavior and there is a whole layer that hasn’t been integrated yet; your phone understands what you’ve done and where you are, but it doesn’t understand how you feel and it won’t ever be able to do that with high fidelity.” Microsoft’s Oliver addresses one key issue with AI and desire by considering the following question: “How do you create something that’s going to be desirable? Do you take insights from how people are behaving or do you change that behavior or tap into behavior?” He explains that if an algorithm is working properly, it can predict what you might like and what you won’t like. He says, “Humans are based around familiarity, so we like to see things that are relevant in that space and time, but we also like novelty.” Yet, the potential for AI, as explained by Oliver, is that it will eventually be able to investigate the data and understand the desire path that the brain creates. However, he says, “You don’t want [the desire path] to be predicted, you want it to be existential.”
AI will certainly change the present shape of the advertising industry. However, what won’t change is the ability to reach people in a way that – for now – only fellow humans can. Ads that surprise people and tap into the human condition with emotions like empathy are typically those that work best.
Although there are limitations for AI’s potential in the advertising industry – it’s not replacing creative jobs yet – this certainly doesn’t mean its significance should be overlooked. The benefits of AI for advertising can be as simple as adding value through efficiency and as complex as hyper-individualized targeting through a VPA that knows its audience better than any marketer has ever dreamed. In order for the industry to delve deeper into the potential of AI in advertising, Finley advises, “Imagine what you would like to be possible, because, in a lot of cases, you might come up with things that are impossible, but you also might surprise yourself.” He believes that there will be a competitive advantage for advertisers and marketers who look at AI as early as possible.
Know your audience
It’s no surprise that data will play a major role in AI’s future. Algorithms are being written to collect and aggregate that data more and more effectively and efficiently. Brand-sentiment data will become more important as VPAs develop; this is how marketers – and then devices – will start to know their target audience. And yet, Slavin is not convinced that brand sentiment should hold this much weight.
He says, “The way [machines] will fail may have more to do with intelligence input than output.” The example he uses is Twitter and the amount of emphasis being placed on – and work being done – understanding brand sentiment on the platform. He finds that brand sentiment is too easy to game and he remains skeptical about whether one can understand someone’s sentiments through a declaration of those sentiments.
For instance, “there may be a lot of people who love Ivory soap, but they’re probably not going to tweet about it. They’re probably not going to tweet about most things,” he says. But, he maintains that it is precisely those online declarations of brand sentiment that the industry is looking toward for signals of insights. “It’s not that there is no value in it, but [these signals] have disproportionate weight because they’re quantitative and come from a machine,” he adds. Slavin warns that the numbers provided by brand-sentiment analysis are volatile and, because of the scale, there is no way to determine the mistakes being made.
Nevertheless, whether it is through brand sentiment or another method of data congregation and analysis, VPAs will become as important to the purchase decision as the purchaser.
Shoulders of giants
When the world wants to know what the next best thing in tech is, it’s sometimes useful to see what the big companies are doing. In the case of AI, not one but nearly all the tech giants of today are in a race to bolster their AI initiatives.
Google started early, as Microsoft’s Oliver points out: “The great thing about what Google did – and this is why it changed the industry: it said that when someone searches for something (there is an intent), [they should] see relevant advertising. It should not be based on how much someone paid. The [notion] of relevance is changing the whole industry around programmatic.” Since it perfected search, Google has taken on a series of AI initiatives, like Google Glass and a series of investments in AI start-ups. Google’s in-house team, Quantum AI Lab, is a collaboration between Google, NASA and Universities Space Research Association (USRA), which targets quantum computing to help develop machine learning.
Facebook has been open about the personal incentives driving its interest in AI. For the company, it’s about coming up with better communication solutions and anticipating the needs of users. For example, Facebook is actively developing better algorithms for image recognition. Earlier this year, the company launched Facebook Moments, which is a separate application that uses facial recognition technology developed by Facebook AI Research (FAIR) to help people organize photos privately and share them with friends. One of the most recent progressions in Facebook’s AI mission is the creation of Facebook’s M, a technical assistant built into the Messenger app. Currently being tested by a select audience, M allows users to send a message and get a response about anything from restaurant reservations to actual purchases. As users interact with the app, M becomes more familiar with their individual behavior. M is an example of how machines and humans can work together to create a more efficient solution for users, as it is not solely AI-fueled. Actual Facebook staffers are also working behind the scenes to ensure every request receives a response.
Apple is also quickly scooping up AI companies, an effort that resulted in the acquisition of two AI start-ups in just four days last month. Its latest acquisition, Perceptio, is a company that specializes in an image-recognition technology for smartphones. The other company, UK-based VocalIQ, is a start-up dedicated to the development of a technology that helps computers understand human speech. Additionally, according to Reuters, Apple plans to hire 86 full-time AI experts. Siri is Apple’s equivalent to a VPA, who has been on the market for quite some time.
Earlier this summer, Twitter began hiring for a new AI program called Cortex. However, according to a July 2015 Business Insider article, the job vacancies provide the only clue to this new initiative. One goal, according to the vacancy announcements, refers to the development of automatic content understanding. The announcement goes on to highlight the limitations of manually defining features to represent the data collected on Twitter and that “Twitter Cortex is responsible for building the representation layer for all this content.” Last summer, the company also acquired AI start-up Madbits, whose founders explain their technology as follows: “Visual intelligence technology that automatically understands, organizes and extracts relevant information from raw media.” Furthermore, in June this year, it also acquired Whetlab, another AI start-up with technology that helps automate and, therefore, speed up machine-learning experiments.
For better or worse
While it’s only natural to get excited by new technology, it may be worth taking a moment to consider its downfalls as well. The most basic fear surrounding AI is the effect it will have on certain professional industries. One purpose of AI is to replace functions that, at one point, came from human intelligence. This means a machine will replace people who were once getting paid to do something.
Some jobs, as MIT’s Slavin pointed out, like truck driving will actually be in danger of extinction because of AI, but as for the creative industries, most aren’t concerned. Annalect’s Syed says, “We still don’t really understand what moves the human mind; humans don’t understand it, machines sure don’t understand it. We know what a very narrow part of a purchase decision looks like because we can get data back from click-streams. But everything that leads up to that is the art of it and that doesn’t change.”
He adds that, although there is a science and an art to everything – and that science will be dominated by intelligence in the near future – the artistic side of the industry will be driven by humans “because we will still push the need for desire. We will understand desire.”
The fact that science and intelligence will dominate the AI revolution within the ad industry has led to the topics of data and privacy becoming part of an ongoing global conversation. Wired writer Finley states the basics: “No one really likes to see the stalker ads that follow you from site to site. We start to realize how much data is being collected and, if it leaks, that can be a huge problem.”
He uses a current event to illustrate how companies are skirting regulations. “It’s so hard to actually know where your data is going, it’s relatively easy for companies to just cheat. Take the Volkswagen emissions scandal (see page 40) [for example]: companies have a pretty big incentive to cheat with software, especially if they think they’re never going to get caught.” The general public is concerned, but experts suggest that it’s because they aren’t yet being made aware of the benefits of sharing that data. Today, there is little immediate correlation between shared data and relevant advertising and people aren’t buying it. Nonetheless, it seems there is still hope for a positive consensus around data sharing. Oliver mentions a study carried out by MIT in Italy where it gave data rights to a small community; brands then came in and explained what these individuals would get in return for their data. “What they found was that when people understood the value exchange, they ended up sharing their data in exactly the same way, if not more so,” he says.
Although people are concerned about privacy and ethical practices surrounding data, there is another threat when it comes to data sharing. It’s like a chicken-and-egg situation, where advertisers can’t serve relevant ads without consumer data and consumers won’t believe in data sharing unless they see relevant ads – or what they find to be a fair value exchange. As Oliver puts it: “You can’t get relevancy without giving off your data, so we’re in this interesting moment where Google, Facebook and businesses like that can’t operate unless they have that data.” Probably worse is the idea that consumers may understand the value exchange and don’t mind sharing their data, but “they’re going to be much more protective if you don’t deliver the experience that they want,” adds Oliver.
The concern around the further development and use of AI does not stop at the spillage of data but also extends to mistakes. AI has been known to make mistakes before, some small and indiscernible and others relatively catastrophic – like the Flash Crash of 2010 for which investigations concluded that both traders and algorithms were to blame for the drop in the market. However, for some, it’s a case of failure being a stepping stone to success.
As Syed says, “It has to make mistakes. [In fact], there is a literal algorithm called error correction, which has to make mistakes before it can correct its error and do better.” And so, it seems, making mistakes is simply part of the machine-learning process. However, he does admit that now internal stress tests are performed on algorithms to allow the corrections to take place before they are served to users. Finley explains that AI’s limitation currently “is either having to trust it blindly or having humans in the middle to spot-check its work, which doesn’t quite defeat the purpose, but does make it less efficient.”
And efficiency, after all, is a huge part of AI’s intent. “You want to depend on less human intervention and to fully rely on [machines] and not have any humans in the loop, but right now you have to have people checking the machine’s work,” says Finley, which might seem counter-productive. But the more advanced AI and algorithms get, the more difficult it will be for humans to look over machines’ shoulders.
Strength in numbers
Many individuals who are very qualified in the field of AI fear the possibilities of this technology in the future. Stephen Hawking has given strong warnings against letting it go too far too quickly. He has said, “The development of full Artificial Intelligence could spell the end of the human race. It would take off on its own and redesign itself at an ever-increasing rate. Humans, who are limited by slow biological evolution, couldn’t compete and would be superseded.”
However, there are many experts in the industry who don’t feel that this is a real threat – at least not yet – and then there are some who are neither here nor there. “I’m in the middle here. We are still really far from having machines that program themselves and do the things it will take for them to spiral out of control. Our progress is so incremental, [it] gives us the ability to control how it ends,” says Finley. He uses the example of the stock-market Flash Crash to illustrate how royally AI can screw up and, although AI didn’t take over the world, these mistakes are a big issue. Finley explains that most people who worry about the apocalyptic AI scenario are forgetting one crucial detail about how it is being developed today: “Most of these AI machines are going to be very purpose-built and they’re not going to be able to do a whole lot of different stuff. Even as you start to make them smarter and smarter, they’re not necessarily going to want to do anything else because they’ve been programmed to accomplish one particular thing.” However, “that might not always be the case, but we’re designing them to give us particular information or do particular tasks, not to be some sort of overlord that can do everything,” he adds. He uses an example of war and AI developing the capacity to launch missiles. “People who are worried about these things aren’t really thinking about how human politics really work and the way humans want to be involved with the decisions that effect them,” he says. He explains that this thought process is assuming that “if something is better or more efficient, then we’re going to use it and that’s not really the case.”
This fear is only normal when it comes to the unknown. Syed says, “It’s like when the Internet first went mainstream and people got the very first virus, people become fearful of what they don’t really understand.” But, he notes, people now use the Internet every day; they even buy devices and cars with the Internet built in. He continues, “People adopt it regardless of the fear, because, if they live their life one day at a time, that fear won’t actually exist until something grand happens.” Besides, these extreme examples are just in the movies, he says. Oliver rationalizes the fear of robots taking over by saying, “Our brains can only make one decision at [a] time, so people are freaking [out] about robots beacuse we don’t know what can happen when a computer can make multiple decisions.” That’s when “we go beyond human capacity, but we’re [far] away from that,” he adds.
Part of understanding the current state of AI is remembering that humans are actually behind it. During his talk at BrainScape, Slavin reminded the audience that “as we build systems that determine [everything from] fraud to pleasure, we forget that there was an idea inside there from a human”. In his talk, he highlighted the vast difference between how a human sees and how a computer sees things.
The main message of his talk – an idea that has been adopted by many AI professionals – was that, right now, Artificial Intelligence has no capacity without human engagement. As he said, “AI gives us answers but not explanations.” He noted, however, that the issue lies in the fact that answers only succeed or fail. Meanwhile, Slavin emphasized the importance of collective intelligence – that of man and machine.
In advertising, it seems that jobs will certainly not be lost on the count of AI. Instead, insights will be gained, for example, from data that humans cannot possibly sift through. However, machines cannot replace the stories that are created by humans for humans. And that’s why Finley says, “Pairing humans with machines will always be the better bet.”