5 Lessons Learned By Working At An AI Startup
By Pauliina Alanen. This article has been previously shared on Pauliina’s personal blog.
There it is, the most misunderstood term of our time: artificial intelligence (AI). It is the single most hyped technology today that means everything and nothing.
Four companies in ten get described as an AI company without them being one. Reasons for this might include confusing plans with the present, seeming more advanced than they actually are, and even just by attracting VC money.
For Pauliina Alanen, Marketing & Communications Lead at Silo.AI, the company she works for does not fall into that category. In fact, Silo.AI is full of hardcore Machine Learning(ML), Natural Language Processing (NLP), and Computer Vision (CV) researchers, most of them with PhDs.
As a Marketing & Communications Lead at Silo.AI, Pauliina has dug deep into the field and works towards making this unknown territory more approachable to everyone outside the circle. She understands that not everyone is a Data Scientist but she believes everyone would benefit from understanding more about AI.
Here is what Pauliina has learned for the past 6 months at Silo.AI:
1. It’s OK for AI to be boring
When we build AI solutions, we are building them as “part of our clients’ existing workflows”. What does that mean? Well, the technical side focuses on the machine learning model and getting a sufficient amount of detailed data. First, we build a model, train it with training data, and then we let the model interact with the actual data and wait for the outcome. It sounds simple, but this technical part is a big part and it is only one part of an AI solution.
Business wise, “being part of a workflow” means that we are not disrupting everything the client has (they would never buy it), but instead, trying to build an AI solution as part of their already existing processes. We need to go to “business-problem first”.
In a case we had, our machine learning tool to predict flight delays became just another data input to the airline’s optimization system. Doesn’t sound super exciting at first (a several months’ project turns into a small button in the huge dashboard), but when you think about it, this tool can lead to a big impact. Better prepared delays equal cost savings for the company and happier passengers.
Equally boring may seem one of our computer vision quality control systems: the solution goes through tons of sewage pipe video material and flags anomalies in the wastewater. You can’t help to think “Is AI taking our jobs to stare at the surveillance tape for crap flow? I would not want any human to waste time in that.
2. Any business function can have an AI solution
Some business functions are inherently data-driven. Any number of heavy area where calculations and estimates are part of the day-to-day work is a match made in heaven for AI. What I have realized through my job is that any business function from HR, supply chain to legal can and should use AI. Even HR and People Operations.
As your grandma used to say: Data is oil for artificial intelligence.
When you ask people to describe their jobs, they give you a few sentences about the tasks they enjoy the most. They probably won’t list the endless updating of reports, drafting contracts, scheduling meetings, and scrolling through emails to remember what tomorrow’s workshop is about. These repetitive small tasks take a lot of time in any department, and they could all be given to AI. That way, people would have more time for interesting and challenging human-suited work.
3. Finding the right problem is difficult
Although using AI in all the departments for the smallest impact would be beneficial, clients won’t just go for any AI solution that they could benefit from. The main reason is due to time, budget and reality constraints. Clients need to pick the area where the solution will have the most impact, a budget can be found and there is an urgent fire to put off.
Finding this type of problem takes time. My company has found it necessary to create Prestudy, Design Sprint and PoC-like frameworks in order to discover these opportunities faster and more efficiently. It is hard to know what is possible and where to start. Designing business driven AI solutions is (at least in my eyes) as hard as getting the model right.
4. Artificial intelligence can save us
With something as powerful as machine learning, you can potentially solve extremely complex problems. Such as climate change.
I believe that technology can help us become better. It doesn’t mean that we can drop everything else we are doing both as a consumer (reducing flight time, recycling) and as businesses in our work, but as DeepMind has shown, making things more efficiently can save astonishing amounts of energy.
This power is one of the reasons why I joined an AI company, and why so many people are so excited about AI.
Because of this immense power, it is crucial to make sure AI efforts stand on good ethical ground. It is scary, the degree of real-looking fake videos you can create to support your propaganda and to brainwash even the smartest of people. However, AI can also be used to fight this scam and to bring in metadata about the photos and videos online that otherwise would be near to impossible to interpret as untrue.
5. You don’t need to code or build models, but understand what they can do
As I’m sure you’re aware, constant learning is the key in this fast-paced world. This has definitely become clear in the AI industry.
Forget the endless categorizing into technical and non-technical people. AI is not too difficult to understand, once you get someone who speaks your language to explain it to you. Accepting you will always, only grasp a limited amount at once, it will help you (the basics of any learning process).
To end this chat and as a personal recommendation from Pauliina, start by taking the Elements of AI course. The course has enlightened more than 150 000 people all over the world. Then, continue by watching videos such as CGP Grey’s: How machines learn or read Tim Urban’s: Wait but Why posts on AI (part 1, part 2). If you want to go further, take the infamous Andrew Ng ML course. Simply put, google it and learn. That’s what machine learning experts/software developers/most adults do all the time. Last, talk about this stuff with your friends, colleagues, and family. Understand and discover together.
Maria 01 is a startup community campus in the heart of Helsinki. We are growing to be Europe’s largest startups campus with 70 000m2 of space for over 5000 tech-heads. Subscribe here to get our news!
