From scheduling meetings, creating movie thumbnails, responding to emails to analysing data, these are novelty applications of AI that look to change the fabric of future work.
The one theme that runs across all these applications is how AI can be used to automate tasks that are laborious if not outright boring if done manually.
How would you teach a program to recognise cats? You could try and do this by programming in explicit rules like “cats have pointy ears” and “cats are furry.”
But what would the program do when you show it a picture of a tiger?
This method poses 2 challenges:
1. Programming in every rule required would be time-consuming.
2. Describing characteristics such as furries or pointiness would be difficult.
Another (and better way) to achieve this task is by letting the machine teach itself: you give it a huge collection of cat photos, and it looks through those to find its own patterns in what it sees. It connects the dots, pretty much randomly at first, and in time, it gets pretty good at saying what is and isn’t a cat.
So in short, AI is a piece of software and like many other types of software, can take some input information, and turn it into some other information, the output.
The only real difference from many other bits of software is that the programmer didn’t have to give it step by step instructions on how to do that transformation, and might not even know what those steps are.
AI powered capabilities for the workplace
Here’s what doesn’t make for a great AI app: taking something we already can do easily, like ordering pizza, and just slamming a chatbot on it.
The real power of AI is augmenting and enabling humans. A great AI app will not just help users do something faster, but enable them to do something they couldn’t do before. This includes things users couldn’t afford before: AI will empower all of us to have our personal assistant and support staffs.
1. AI powered capabilities to help with scheduling meetings
If you’ve been scheduling meetings manually, one at a time, the old-fashioned way of exchanging emails and availabilities, then chances are you’ve been wasting hours every week for this mundane task.
On average, I have 5 meetings a day. Even if it just takes me 10 minutes to schedule one (it takes more, plus the distraction from email ping pong), every week it takes me approximately 4 hours to just lock in time.
X.ai is here to solve just that. You can use a wide range of tools (some are AI Powered, some are tools with great UX) to minimise the amount of time and wasted effort to ‘lock in a time’.
2. AI powered capabilities to help with analysing data
While there are a lot of buzz about amazing machine learning in the last few years, as a data analyst, I can wholeheartedly say that I spent more than 80% of my time doing simple, almost too repetitive-by-nature tasks.
Whenever someone asks for a report on how sales went last week or a new dashboard so we can visualise our product revenue by a new attribute, a part of me died because I knew I’d be wasting the next 4-5 hours doing something that is probably looked at once. (A lot of people think dashboards, once built, can be used on a regular basis and need no maintenance. While this is true for maybe 2-3 dashboards in an organisation, the hundreds rest suffer the fate of being looked at once and never again)
Hyper Anna is here to solve just that, to enable everyone to explore business insights instantly without the need of an analyst spending hours to create a custom report or dashboard.
This does not just save analysts hours in a week but fundamentally enables everyone in the workplace to become more insight literate and to make better decisions.
Building AI powered products
The below example of creating movie thumbnails might not be all the way work-related but it is hopefully relatable for most. It is also an excellent example of how automation & personalisation are the key theme in AI powered functionalities or products.
Over the years, Netflix has been introducing fantastic examples of how products can leverage AI to provide superior values that can not be done by any other way and do it in a way such that tech isn’t in your face but blended into the experience.
Netflix application (algorithms) suggests personalised movie thumbnails based on user’s preferences and liking.
For the same Good Will Hunting movie below, one user identified as a comedy fan would be shown a Robin Williams (comedian) thumbnail, whereas another user identified as a romantic comedy fan would be shown a kissing thumbnail featuring Matt Damon and Minnie Driver. While not perfect, Netflix’s algorithms suggest that such level of personalisation based on user profile characteristics increases probability of click thru rates.
What data does Netflix use to create these personalised thumbnails / artwork?
- A 1 hour episode of Stranger Things has >86,000 static video frames
- These video frames can each individually be assigned certain attributes that are later used to filter down to the best thumbnail candidates through a set of tools and algorithms called Aesthetic Visual Analysis (AVA). This is designed to find the best custom thumbnail image out of every static frame of the video
- Netflix Annotation — Netflix creates meta data for each frame including brightness (.67), number of of faces (3) , skin tones (.2), probability of nudity (.03), level of motion blur (4), symmetry (.4)
- Netflix Image Ranking — Netflix uses the meta data from above to pick out specific images that are highest quality (good lighting, no motion blur, probably contains some face shot of major characters from a decent angle, don’t contain unauthorised branded content, etc) and most clickable
Then based on users’ viewing history & attributes, the ‘most recommended’ thumbnail is used for display.
With this method, each content can be marketed towards users in a way that traditional methods simply can’t.
Promotional images taken from four different Netflix accounts for the movie “Set It Up.” In this case, depends on the user’s background (anglo, hispanic, asian, african-american), a different movie thumbnail is shown.
Taking this too far however, and the suggestions could be misleading in terms of a thumbnail accurately representing that movie.
Of course, this algorithm will likely be fine-tuned over time, but the lesson here is don’t overdo it when capitalising on data — apply some common sense to balance it out.
While building an AI product is generally complex, once you break down your problem statement into machine learning problem(s), implementing some (or all) of them can be more achievable than you imagine.
Broadly speaking, you can split AI into two categories:
1. Ones that do tasks so common they are needed in many different contexts, like transcribing voice into text.
2. Ones that tackle more unique tasks, like detecting defaulting rate on loan applications.
The distinction is important because the common problems are often already solved, and you can use an existing AI rather than creating your own one.
Here is an example to help illustrate this. If my task is to sort product photos based on category, I can use an existing AI to do this easily.
But if my tasks is to analyse thousands graphs and identify patterns, an existing AI might not do so well. This is when building a custom AI with my own datasets would give my product a much higher accuracy.
You might ask why identifying a hat is important, imagine if you are a retailer and you need to tag your thousands of products by different attributes, doing it manually would take an enormous amount of time and what about the updating new products or introducing new attributes?
This is where the above simple execution of algorithm can provide a substantial improvement in efficiency and productivity (and also work satisfaction).