Is AI going to revolutionise the events industry?

AI is a very interesting tool, and one that’s making a huge impact on many sectors, but contrary to some of the things I’ve read, I can’t currently see how it will have any big impact on events.

The most common form of AI that is being talked about in our industry at the moment is “Machine Learning” (or ML). In machine learning, as its name suggests, the machine learns how to figure things out - this is how systems have been created that can classify the photographs that we’ve taken and that have been uploaded to one of the photo platforms. Go to Google Photos and search for “flower”, “cat”, “dog” or “Mercedes” (or whatever keywords will work for your photos). You may be surprised by how accurate these things are!

In ML there are two forms of training and one or the other is required for the machine to learn something - these are called supervised and unsupervised learning. 

In unsupervised learning you give the machine a set of “things” - they might be data records or photos - and you let the machine group them into sets. These sets will have characteristics that the machine has identified but you won’t necessarily know what they are. The task here is to look at the resulting sets and try to figure out what is common about each set in order to figure out what the machine “saw”. In this sort of ML, there is a high rate of instances where the results aren’t any use and you then reset and start again with a slightly different approach.

With supervised learning - you give the machine a set of “things” but now, you tell it what the desired groupings are too. If we give it 30,000 pictures of cats and dogs and we tell the machine that set A are cats and set B are dogs then the machine will go away and try to learn how to differentiate those sets.

It is important at this point to stress that ML is best at problems like this where it is really hard (if not impossible) to write a set of rules to explain the problem and how to solve it. Think about it - you couldn’t come up with a list of characteristics that would define what differentiates a picture of a cat from a picture of a dog yet you can easily look at a photo and tell which is which.

Another really important fact about ML is that you won’t be able to look inside the resulting process and necessarily know what it is doing. If you give it a photo that is a cat or a dog and it identifies it correctly, you won’t necessarily be able to understand exactly HOW it did that but if the result is correct enough of the time then that isn’t really a problem - in that situation anyway.

Let’s take a look at a specific example of a potential “match making” process and why AI can’t do it. The example is fairly simple and will probably be familiar to most of us in the events industry. I often stay in hotels and I tend to use one or two online platforms to actually book my stays.

This means that both these platforms have hundreds of records of my hotel bookings and yet, they couldn’t possibly identify why I booked any of them or accurately predict which hotels I might be likely to book in future because there are simply too many variables that the platforms have no concept of. My choices could be based on my knowledge of the hotel’s air conditioning (or lack thereof), the height restrictions of the car park, whether I’m driving or getting the train, whether I’m going to a meeting later in the morning or working onsite at a venue until silly o’clock, whether I just fancy trying a new hotel or a specific one, whether it’s for work or pleasure, or even whether it’s close to that restaurant that I like and dozens of other reasons. AI could never know all of those factors because there are too many personal reasons for my choices.

These booking sites take millions of bookings every year and will have gathered vast quantities of data and have enormous IT budgets but they can’t choose a hotel for me because they don’t know enough about me and my motivations for this particular booking.

Bear in mind that these sites generally ask how my stay was and often get my feedback on what I did and didn’t like but they can never ask enough questions to truly understand why I made that booking - mainly because I’d then be bored of the survey and wouldn’t complete it!

So how does that apply to the idea of AI automatically making appointments for people at events? Well, I’ve seen some people claiming that they have “AI platforms” that can automatically match make and create appointments for attendees but I don’t believe this. 

Think about the last exhibition you attended - you might be able to explain to another person some of the reasons why you do or don’t want to meet with a particular exhibitor but that might be as simple as “you’ve worked with them before and don’t want to again” or “they have a reputation that you don’t like” or “you already know and work with them” or plenty of other reasons. Can the alleged “AI platform” mine this information from your provided registration data? No. I’ve even seen one platform claim to mine your social media feeds for your preferences which sounds fairly horrific from a data privacy standpoint but even then it certainly wouldn’t be able to infer anything useful about who i want to meet from my social media accounts.

Even if we assume for a minute that these AI algorithms could find all of this data. To be able to train it it would need to know if its selections were correct, so we would need to tell it the result of tens of thousands of meetings that had been scheduled. The trouble is, nobody can know whether it was actually a good meeting straight afterwards (unless the potential customer placed an order there and then!). It might have seemed positive but then come to nothing - alternatively, I might only find out that it was an amazing meeting 18 months later when a huge order comes in. That information never filters back to the organiser in my experience so there are no metrics for what constituted a “successful meeting” and so therefore the AI cannot learn. 

Facial or voice recognition are slightly different things and may see some good results in our industry because those models have been trained in all sorts of other industries with millions of records and so they should work well. For other AI tools, unless we can find AI algorithms that have been trained outside of the events industry that will work without any changes then I doubt it’ll happen. I’d love to hear from people who are really using AI in anger and can explain to me how they’re training those ML networks. 

As it stands, I suspect that those claiming to use AI in our industry are just using the term as a marketing buzz word – they aren’t really using AI because they just don’t have the data it needs but I’d love to be proved wrong.