PUBLISHED
31 October 2016
When new trends and technologies burst onto the marketing scene,there’s always a frantic effort to either keep up or provide guidance, especially when serious amounts of money are involved. It happened with social media, it happened with personalisation and big data, and it’s happening now with artificial intelligence.
We’re approaching the top of the hype cycle where, like teenage sex,everyone is talking about it but very few are actually doing it. Conditionsare perfect for the snake oil salesmen to move in. But there’s realsubstance behind some of work being done in this field, and in this post I’lltry to navigate through the fog of rhetoric to understand what’s requiredto make the most of the significant opportunities.
Back to the future
To go forward, let’s first look backwards. AI has been a popular subject inscience fiction for decades, often running alongside robotics to createsome of the most original storylines of all time. These stretch from theutopian (Star Trek, Her and I, Robot*) to the dystopian (BladeRunner, 2001, The Matrix and The Terminator), and everywhere inbetween. As a result, the term has become attached to the future, framingexpectations around the art of the possible in the present.
As with all science fiction, it serves as inspiration for the people working onreal-life applications. We still have a way to go before we catch up withfiction, but actual developments have, in their own way, been no less notable. The Bombe that helped to decipher Enigma in WW2 meets the definition ofAI, and it wasn’t the first of its kind. Machine learning has been aroundsince the 1950s, growing alongside the development of computers, andheld back mainly by computational power. More recently, as processorgrowth has followed Moore’s Law, the high-profile milestones have comethicker and faster, from Deep Blue beating Garry Kasparow in 1997 toWatson winning Jeopardy (and now trying to cure cancer), to Google’sDeepMind beating Lee Sedol at Go, through to Facebook’s Bots unveiled at F8.
When developments in the real world start to mimic science fiction, thisgets a lot of attention. Apocryphal headline grabbers such as the MicrosoftBarbie Bot will gradually be replaced by the exceptions-that-prove-the-rule,such as the recent incident with Google’s self driving car (a.k.a. JohnnyCabs).
But the really big advances will come when we crack somethingcalled general artificial intelligence.Experts concur we’re many years awayfrom that, but in the meantime we can expect an array of intriguingdevelopments.
There’s a clamour of investment taking place, usually a good sign of firebehind the smoke. From 2 per cent of VC investment in 2013 to 5 per centin 2015, interest is significant and growing. Google, Facebook, IBM, Apple,Salesforce, Cisco, Intel and many more have invested hundreds of millionsin this space over several years, particularly in the adtech and financefields. Our parent group, Publicis.Sapient, has bought a minority stakein Lucid.ai. Wired luminary and all round good guy, Kevin Kelly, says: “Thebusiness plans of the next 10,000 startups are easy to forecast: Take X andadd AI”.
To go forward, let’s first look backwards. AI has been a popular subject inscience fiction for decades, often running alongside robotics to createsome of the most original storylines of all time. These stretch from theutopian (Star Trek, Her and I, Robot*) to the dystopian (BladeRunner, 2001, The Matrix and The Terminator), and everywhere inbetween. As a result, the term has become attached to the future, framingexpectations around the art of the possible in the present.
As with all science fiction, it serves as inspiration for the people working onreal-life applications. We still have a way to go before we catch up withfiction, but actual developments have, in their own way, been no less notable. The Bombe that helped to decipher Enigma in WW2 meets the definition ofAI, and it wasn’t the first of its kind. Machine learning has been aroundsince the 1950s, growing alongside the development of computers, andheld back mainly by computational power. More recently, as processorgrowth has followed Moore’s Law, the high-profile milestones have comethicker and faster, from Deep Blue beating Garry Kasparow in 1997 toWatson winning Jeopardy (and now trying to cure cancer), to Google’sDeepMind beating Lee Sedol at Go, through to Facebook’s Bots unveiled at F8.
When developments in the real world start to mimic science fiction, thisgets a lot of attention. Apocryphal headline grabbers such as the MicrosoftBarbie Bot will gradually be replaced by the exceptions-that-prove-the-rule,such as the recent incident with Google’s self driving car (a.k.a. JohnnyCabs).
But the really big advances will come when we crack somethingcalled general artificial intelligence.Experts concur we’re many years awayfrom that, but in the meantime we can expect an array of intriguingdevelopments.
There’s a clamour of investment taking place, usually a good sign of firebehind the smoke. From 2 per cent of VC investment in 2013 to 5 per centin 2015, interest is significant and growing. Google, Facebook, IBM, Apple,Salesforce, Cisco, Intel and many more have invested hundreds of millionsin this space over several years, particularly in the adtech and financefields. Our parent group, Publicis.Sapient, has bought a minority stakein Lucid.ai. Wired luminary and all round good guy, Kevin Kelly, says: “Thebusiness plans of the next 10,000 startups are easy to forecast: Take X andadd AI”.
What makes it AI
So let’s take a step back and ask: What is AI? The dictionary definitionsuggests it’s a broad church. Arguably, a simple calculator qualifies asartificial intelligence. It’s not surprising then startups, vendors andsuppliers claim their products and services include an element of AI. Lookout for wagons being hitched to that horse.
One way to start conversations with vendors, partners and potentialrecruits and employers is by asking them what they mean by the termartificial intelligence. As with big data and innovation, it means different things to different people, and time can be wasted assuming you’re on the same page. We’ll need to develop our lexicon before we can sort the wheatfrom the chaff. In the marketing and technology sector, this territory is currentlydominated by the data scientists. The challenge of extracting value fromhuge data sets is in some ways fuelling the interest in AI. The goal here is tomake even better use of data to support strategic planning and drive real-time decision making, reducing our dependency on expensive, fallible datascientists and customer support staff, and increasingly automating the nextsteps without human intervention. There is already a wide range of examples in this space, from automatedpricing to predictive customer care, personalisation of ad targeting, andmore. This list from econsultancy of 15 examples is worth a scan. And as time goes on, we can expect these services to become moreaccessible to less technically minded people. There has also been plenty oftalk about bots, especially after Facebook introduced a way of embeddingbots in its Messenger app. Facebook refers to it as a hybrid betweenlanguage recognition, decision tree mapping and customer care – in otherwords, it’s essentially a form of Interactive Voice Response for text. For the best examples, check out Poncho and either CNN or the Wall StreetJournal on Messenger. The demos look good, but early results are poor.And even when it improves, this only barely qualifies as AI. Until it reachesa greater level of sophistication, it’s mostly creating the illusion ofintelligence because of its dependence on pre-determined decision trees. That isn’t to say the bot approach isn’t worth exploring; far from it. Justdon’t get lulled into thinking you’re in the vanguard of artificial intelligencewhile you’re at it, and get ready for a frustrating time in the short term. While general artificial intelligence might be many years away, it’s worthkeeping an eye on emerging developments in that space as well. They’llprobably be useful well before they hit their stated goals.
One of the more interesting developments is viv.ai, a service beingdeveloped by one of the ambitious Siri co-creators, Dag Kittlaus. Not onlydoes he want to create a cloud-based platform for finding connectionsbetween disparate data sets, he wants to put a universally recognised voiceinterface on it. He’d like his ‘V’ logo to be as ubiquitous as the bluetoothlogo, so we know how to engage with the system. Apparently we speak 3-4times faster than we write, so this makes sense so long as the system isn’t plagued by the same challenges as Siri or Google Now. When it comes to voice interaction in general, the early signs fromAlexa (from Amazon) are very promising, so maybe we’re on the cusp of abreakthrough here.
So let’s take a step back and ask: What is AI? The dictionary definitionsuggests it’s a broad church. Arguably, a simple calculator qualifies asartificial intelligence. It’s not surprising then startups, vendors andsuppliers claim their products and services include an element of AI. Lookout for wagons being hitched to that horse.
One way to start conversations with vendors, partners and potentialrecruits and employers is by asking them what they mean by the termartificial intelligence. As with big data and innovation, it means different things to different people, and time can be wasted assuming you’re on the same page. We’ll need to develop our lexicon before we can sort the wheatfrom the chaff. In the marketing and technology sector, this territory is currentlydominated by the data scientists. The challenge of extracting value fromhuge data sets is in some ways fuelling the interest in AI. The goal here is tomake even better use of data to support strategic planning and drive real-time decision making, reducing our dependency on expensive, fallible datascientists and customer support staff, and increasingly automating the nextsteps without human intervention. There is already a wide range of examples in this space, from automatedpricing to predictive customer care, personalisation of ad targeting, andmore. This list from econsultancy of 15 examples is worth a scan. And as time goes on, we can expect these services to become moreaccessible to less technically minded people. There has also been plenty oftalk about bots, especially after Facebook introduced a way of embeddingbots in its Messenger app. Facebook refers to it as a hybrid betweenlanguage recognition, decision tree mapping and customer care – in otherwords, it’s essentially a form of Interactive Voice Response for text. For the best examples, check out Poncho and either CNN or the Wall StreetJournal on Messenger. The demos look good, but early results are poor.And even when it improves, this only barely qualifies as AI. Until it reachesa greater level of sophistication, it’s mostly creating the illusion ofintelligence because of its dependence on pre-determined decision trees. That isn’t to say the bot approach isn’t worth exploring; far from it. Justdon’t get lulled into thinking you’re in the vanguard of artificial intelligencewhile you’re at it, and get ready for a frustrating time in the short term. While general artificial intelligence might be many years away, it’s worthkeeping an eye on emerging developments in that space as well. They’llprobably be useful well before they hit their stated goals.
One of the more interesting developments is viv.ai, a service beingdeveloped by one of the ambitious Siri co-creators, Dag Kittlaus. Not onlydoes he want to create a cloud-based platform for finding connectionsbetween disparate data sets, he wants to put a universally recognised voiceinterface on it. He’d like his ‘V’ logo to be as ubiquitous as the bluetoothlogo, so we know how to engage with the system. Apparently we speak 3-4times faster than we write, so this makes sense so long as the system isn’t plagued by the same challenges as Siri or Google Now. When it comes to voice interaction in general, the early signs fromAlexa (from Amazon) are very promising, so maybe we’re on the cusp of abreakthrough here.
AI and unexpected problem solving
The broad consensus is that we shouldn’t think of AI as simply improvingthe tasks we currently perform, either. We should also think aboutapproaching problems in completely different and unexpected ways toachieve much greater outcomes. AlphaGo’s behaviour is a great earlyexample of this.
“Reinforcement learning”, or teaching computers how to learn forthemselves, appears to be a more fruitful approach than “human teaching”or decision tree mapping, as it’sless constrained by our own humanshortcomings. This is also where neural networks, such as this interestingexperiment from TensorFlow, come in. In their own words, the network isasked to solve a problem, which it attempts to do over and over, each timestrengthening the connections that lead to success and diminishing thosethat lead to failure. It’s a general principle that applies to one of the mostinteresting areas of AI.
What all of this says is we don’t know exactly where the emergence ofartificial intelligence will take us, which makes the whole venture soexciting. Certain aspects, such as those relating to general artificialintelligence, will be exponential by their very nature. That’s when thingsget scary very quickly. For a taste of that I highly recommend this Wait ButWhy article. When people like Elon Musk and Stephen Hawking saysthere’s a significant danger, it’s time for all of us to sit up and take notice.
In the short term though, we can start looking for opportunities to exploit AI technologies as they mature and generate new forms of value. It’simportant to get an early, solid understanding of how the opportunity canbe exploited and, as with the technology waves that came before, a fewwell-chosen bets may pay off handsomely. She who dares wins, but let the buyer beware.
The broad consensus is that we shouldn’t think of AI as simply improvingthe tasks we currently perform, either. We should also think aboutapproaching problems in completely different and unexpected ways toachieve much greater outcomes. AlphaGo’s behaviour is a great earlyexample of this.
“Reinforcement learning”, or teaching computers how to learn forthemselves, appears to be a more fruitful approach than “human teaching”or decision tree mapping, as it’sless constrained by our own humanshortcomings. This is also where neural networks, such as this interestingexperiment from TensorFlow, come in. In their own words, the network isasked to solve a problem, which it attempts to do over and over, each timestrengthening the connections that lead to success and diminishing thosethat lead to failure. It’s a general principle that applies to one of the mostinteresting areas of AI.
What all of this says is we don’t know exactly where the emergence ofartificial intelligence will take us, which makes the whole venture soexciting. Certain aspects, such as those relating to general artificialintelligence, will be exponential by their very nature. That’s when thingsget scary very quickly. For a taste of that I highly recommend this Wait ButWhy article. When people like Elon Musk and Stephen Hawking saysthere’s a significant danger, it’s time for all of us to sit up and take notice.
In the short term though, we can start looking for opportunities to exploit AI technologies as they mature and generate new forms of value. It’simportant to get an early, solid understanding of how the opportunity canbe exploited and, as with the technology waves that came before, a fewwell-chosen bets may pay off handsomely. She who dares wins, but let the buyer beware.