Marketing and market research share common ground: both are obsessed with using terminology that outsiders find difficult to understand.
Is it any wonder, when their definitions are so widely misrepresented?
Are buzzwords for the unimaginative?
You might have heard of ‘influencer marketing’ (advertising through popular social profiles) and ‘advertainment’ (ads that double as entertainment). We’ve been using terms such as ‘gamification’ and ‘millennials’, but industry professionals fall foul by referring to ordinary survey techniques as ‘gamified experiences’, or by describing anyone from thirteen to thirty as a ‘millennial’.
It’s with overuse that these ill-defined buzzwords stray from their true meanings.
We can throw the term ‘AI’ around in popular culture, on the news, and at conferences, but not very many of us know what’s really meant by it. Often it’s used to refer to other complex algorithms that don’t qualify as AI at all.
So how should we define it?
What’s the difference between ‘Artificial Intelligence’ and ‘Machine Learning’?
Sometimes AI is used interchangeably with another term – ‘machine learning’ – but they’re not quite the same thing. In fact, ML is considered a particularly innovative subset of AI.
A concept since the 1950s, AI is a broad term that refers to any technology, machine, or mechanism that’s designed to act intelligently or mimic an intelligent human action.
- Specific AI can achieve specific tasks or skills unhindered, such as playing Go, trading, or dispensing goods (Amazon’s drones or robot hoteliers are good examples).
- General AI is much harder to find as it’s so flexible, capable of solving problems and completing a wide range of tasks as skillfully as humans (so far this is mostly hypothetical).
Traditional 1980s style AI (specific AI) acts as a tree diagram or flow diagram. A computer asks a series of questions and chooses the best next question based on whatever answer it receives.
Here’s an example of your average flowchart. Eventually, it reaches a conclusion (or a series of conclusions) originally mapped by an expert in the field. It’s intelligent in the sense that this arrival seems deeply personal, but in truth, anybody experiencing the exact same pattern will receive an identical response.
So, automated analysis in market research is technically a specific AI, applying this thinking to each statement. Every sentence has been through the above processes and reached a final ‘narrative’. This narrative might change depending on the other sentences that surround it, but ultimately, we’re left with a written report fuelled by experts with Artificial Intelligence.
Simply put, machine learning involves a computer educating itself with use of the data fed to it (often a lot of data).
At Zappi, we use the term to mean that we are trying to teach the machine. This method is the opposite way round to AI, in that it starts at the bottom of the tree diagram, in the data itself, and works its way up. The computer searches tremendous amounts of base data and asks an expert if any patterns it finds mean anything specific. If they do, the expert will teach the computer to recognize these for next time – but by the same token, many patterns might not mean anything at all.
This is why machine learning is sometimes labeled ‘dumb AI’, as for all intents and purposes, it’s an unintelligent pattern recognition algorithm.
It can get cleverer though; what we call ‘neural nets’ are essentially much more complex machine learning algorithms that accrue mass data patterns to predict human behavior. You’re likely familiar with this kind of deep learning as it’s used to introduce new music in Spotify, suggest films for Netflix, and recommend products on Amazon.
Are we close to developing a general, sentient AI?
With Sophia in the news, the humanoid robot that’s recently been granted Saudi Arabian citizenship, it’s tempting to believe that we are nearing a general, all-powerful, sentient AI that exceeds even human intelligence. We’re nowhere near.
Sophia is unable to have a continuous, seamless conversation with others – only those conversations she has been pre-programmed to talk about. Sophia is more likely multiple specific AIs working together.
The surest way humanoid robots such as Sophia will achieve an impression of general AI is by merging with voice operated technology such as Alexa (or the like). This is why Amazon is obsessed with data collection – they want to build a way of having a natural language conversation, thinking about all the different types of human problems in a human way. At no point are they trying to create an understanding of what we’re saying per se, they’re simply applying a brute force machine learning technique to obtain tens of millions of potential conversations mapped as a data bank.
What does this kind of technology mean for market research?
AI is currently a small but complicated area in research right now, but it will become universal in two to three years time, so it’s important we talk about it accurately.
To imply that some AI algorithms are unique to one tool or business is simply not true. Most algorithms are written by technologists at companies like Google, and they are typically open sourced. There’s no real IP in machine learning – what sets one business apart from another is its data. Which is why Facebook, Google, and IBM embrace open source AI; they’re collecting our data.
In market research, there is now more than enough data to generate a bird’s eye view of every test submitted to a single platform. Machine learning can be a fully automated part of this process, and users can achieve a true analysis of their research results against industry norms as well as their own brand-specific goals (instead of an ill-defined set of charts and numbers). We will shortly be seeing the emergence of this as a new industry standard: not only research experts but now marketers can innovate, test, and learn for themselves.
Personable, human expertise ‘built-in’.
‘Intelligence Augmentation’ is our new battleground
‘Intelligence Augmentation’ is a term coined by William Ross in his 1956 (and a topic I covered recently in Research World magazine), acknowledging that the idea of AI alone is not as efficient as a combination of human + machine.
Sat nav serves as a basic example: this is a specific AI combined with drivers’ personal knowledge. Sat nav doesn’t dictate one route or forbid the user from waylaying its course – it will generate multiple routes based on decisions made by the human.
From this perspective, we are going to learn most from both directions of AI; the 1980s approach (top down), combined with machine learning algorithms (bottom up). This is by far the most advantageous, cost and time-effective application of AI, because the middle ground combines the best thinking from both machine and human.
Sat nav aside, look at Iron Man.
Iron Man is smart, interesting, engaged, charming, and handsome, but he’s also incredibly powerful, quick, and well armed. Without his cloak, he can’t be both. The man has his own weaknesses, and similarly, the suit isn’t particularly useful on its own. It needs someone to drive.
At Zappi we are trying to make market research iron men and women.
Are you being oversold AI?
It’s an injustice to inflate what is essentially specific AI as overly complex deep learning. Here are just a few tips that’ll help you spot fake AI claims in the MR marketplace:
- Machine Learning is not sentient AI: Decisions made by machine learning are interspersed with human data and human tweaks.
- Automation is not AI: Terms like ‘robot’ or ‘robotic’ allude to the simple automation of a process or basic, applied AI (such as your Roomba).
- AI is not exclusive: No business has its own impossible to understand secret recipe for AI. It’s more likely open source technology that’s fuelling a frenzied race for data.
- Truly general AI makes its own decisions: Humans will take a backseat when true AI materializes, learning from and communicating with other machines by itself
To learn more about the differences between AI and IA, check out the webinar below in which I spoke with Ray Poynter of NewMR on the topic.
Get in touch with us about Zappi.Pro, Zappi’s own intelligent market research platform