This is the eighth in a series of interviews with members of the Machine Commons supplier Collective. Subscribe to the site to be alerted about future posts, or become a partner today!
Sergey Sukhanov runs AI Superior, a machine learning and data science consultancy whom pride themselves on rapid prototyping and a research heavy company ethos.
How has business been through the pandemic?
“In the beginning and after the first lockdown, people were sceptical."
"Investors were slowing investments, businesses unwilling to invest in new projects and technologies."
"New customers were still happening, but we could feel the market became silent for a few months. Nobody knew what to expect.”
“There were so many uncertainties with the virus. Then, once scientists clarified a few things about our future, the market really started up again. Investment in tools and analytics came back.”
What did you use the down time for?
“We used this time to strengthen our competence in promising areas and developed several analytical products/components."
"We were certain, even if the world changes, people will still require data analysis.”
“Machine learning is already impacting so many domains. There’s huge potential for the future.”
Working on anything interesting?
“Generally speaking, we work across many verticals so we try to make sure the analytics we develop are transferable across business domains.”
“We developed a platform that extracts entities and message aspects, i.e. For restaurant aspects like food there are multiple components to an experience: people can like the food but think the service was poor. This enables a much deeper understanding of any customer review.”
“We adapted it to other businesses. So, airlines for example, or any company that deals with customers can use this analytical platform.”
“We also built a computer vision system that can track different road entities – like cars, bicycles and pedestrians – to perform behaviour analysis. It can also capture more granular information, like the make and model of the car. i.e., BMW 3.”
“We noticed that these analytics were required by different businesses, like retail companies, or a shopping mall: it helps them to understand the cars that pass by and therefore estimate the income of the neighbourhood; or ad companies can understand the customers that walk by."
“Additionally, we see interest from companies that are dealing with security and access control.”
Is developing reusable code a deliberate business strategy?
"We’re always trying to increase the production speed of the development team.”
“Whilst it’s not our business model to create reusable modules, there are generic components to every business."
“Once we approach a new use case, we see how its KPIs and data fit to pre-developed modules and then adjust the system, or develop components from scratch if it is something unique and new.”
With two years of core modules developed, are you getting noticeably more efficient?
“Yes and no! Yes, because obviously more problems to tackle, more use cases available for us to unroll existing packages and integrate.”
“On the other hand, there are always state-of-the-art approaches to adopt and new business challenges that require improvements in performance and design.”
Do you spend much time researching in house, to remain at the state of the art?
“At any one point, we have different teams that execute practical work and research work; we try to alternate our experts who do the research and do project work, so everyone has hands on experience as well as the ability to be close to research.”
“This is crucial for our domain since it is not only about the process but also about scientific creativity.”
“Moreover, these ongoing processes help us identify which directions to go (i.e. in computer vision or NLP), then we try to do research that we can then apply to customers’ needs.”
What’s the most state-of-the-art thing now?
“There are so many domains where changes are happening."
"What we feel is really exciting right now is remote sensing and satellite imaging analysis."
"We’re not really able to talk about the technology specifically but there are so many interesting use cases enabled.”
Go on, satellites!? Tell me something.
“The tech is based on satellite or airborne sensors that include not just usual RGB cameras but also hyper- or multispectral cameras that are able to “see” in spectral ranges that are not accessible to our eyes or regular cameras. With this you’re able to sense different materials on earth.”
“This leads to fascinating tasks, from detecting and classifying objects, isolating buildings and trees, but also including a lot of security applications, such as border control.
"There are some environmental use cases, too, like garbage or oil spillage detection.”
Hold on. You’re working on garbage detection from space?!
“Sounds ridiculous, I know, but there are a lot of coastal areas that suffer from tourist activity. The analysis can improve the efficiency of rubbish trucks."
"Or if there is a ship crash, we can track the debris or oil spills.”
Always-on, hyper-spectrum cameras flying undetectably from space… Privacy concerns?
“This is a valid concern for many machine learning projects, as there is so much data collected. One major thing is GDPR compliancy, introduced in 2019."
I can't see Sergey, but I sense a certain posture change. He get's serious. Privacy in machine learning is no joke and not to be taken light heartedly. Especially not when discussing orbital hyper spectral cameras.
"We're always transparent with customers, building explainable algorithms and ensuring these algorithms are not involving private aspects or people's lives."
“Using satellites is no different, for example not allowing the system to detect people or belongings.”
“Generally, for all of the projects, we encrypt the information and erase relations between a person and the data.”
“The data is protected and stored in the right way so there is no leakage or 3rd party access. On the other hand, often we do not need the data of the person. If you do not need it, we do not use it. Obviously!”
A basic tenet of GDPR is that if you do not need a particular piece of info then you are not allowed to store it.
On the topic. As a practitioner who needs to work with data all day – I’m curious, has GDPR negatively impacted business?
“It's true that it requires us to follow processes and these can take more time or require particular tools.”
“But data is *the* critical asset for us. And for our customers often, too. So, we need to make sure that we protect it and have proper processes to deal with it.”
“Like driving a car, you need to follow rules, or you will end up in a crash.”
Anything you find particularly interesting or unique about your work?
“I like data science and machine learning because in this domain you’re always learning something new.”
“It’s not just the field you are focusing on, never just one subject, for example in biology or chemistry you’re researching particular materials."
"In data science and machine learning, you’re dealing with so many different real-life problems, so you learn many different aspects of life!”
“To bring value to customers you need to understand the problem. You need to dig into the data. Sure, not to the same level of comprehension as your customer, but you need to understand what you are looking for and fulfil the customer requirement.”
“This brings a lot of knowledge about our world.”
“I would probably never have dealt with something like industrial machines, or understood the inner workings of air conditioning units."
"Now, since we had a project in those areas, I’m aware of the challenges! What data is generated. What solutions people are searching for.”
“This develops everyone in the company in a positive way, it broadens the mind. Makes us open and more creative.”
And what about machine learning in general?
“Personally, I’ve been involved in the field since 2011. I’ve been actively participating, contributing and adopting the newest developments in the area.”
“It’s fascinating to see the progress. Every year there are new approaches that allow you to do things you never thought were possible before.”
“Take, for example automatic translation. It would never have happened ten years ago. Now it exists with such quality and speed.”
“Volumes of data and new approaches allow for completely new services and products available to the typical consumer."
"Take Amazon or Netflix, you can see how targeted the content is, how relevant the shopping items or movie recommendations are. And obviously, we are still far from our limits.”
Where does the field go from here?
“One of big problems remaining is that human context is difficult to grasp. Context is fundamental to understanding a problem or creating a solution.”
“This is where ML and AI dev will go next: to better understand the environment and context."
"That would unlock automation in many new areas as well as bring completely new experiences to our daily activities.”
Ok. Context. How would ML better understand context?
“Multiple ways, obviously this shouldn’t be invasive or feel there is something disturbing happening, but there are approaches.”
“You can use computer vision, or sensor data analysis, processing data from mics and cameras, or using GPS from your phone, to name a few. Your smart watch might help a lot with that.”
“It’s about understanding where you are and what you’re currently doing. If you’re having dinner, browsing the internet, or waiting for a bus… the user requirements are different."
"The goal is to sense the mood, the circumstances and your feelings; the challenge is to really develop these approaches to extract insight out of the data.”
“Combining all the data that we are generating will help. You might not understand context from a singular person (read: data source), but when you take it in context of millions of people…”
What are the implications for the typical person? (read: consumer)
“Optimizing your life in different ways based on what you like and prioritise."
"I would like to live this life where I can spend more time on activities that I really enjoy and leave routine tasks to machines.”
“I would really like having services adopted to my personal preferences that have clear and robust interfaces to them. I would like to get targeted with appropriate ads and content versus irrelevant things. I would like to have our environments safer and cars safer.”
“I believe many people share these desires.”
//
Personally, I’m on the fence. I either agree with him or think he’s drinking a little too much of the data science cool-aid. I‘m honestly unsure.
Perhaps my hesitation is besides the point; you can’t put the toothpaste back in the tube – this is what machine learning practitioners seem to have simply accepted.
The fact is: we are generating all this data. If it’s not used positively and deliberately by entities working to improve everyone’s lives… well, it’s not like it disappears.
It’s too late to argue about whether we should proceed. Oh, we are proceeding! The only remaining question is how we should proceed.
//
Last question. Tell me about the future.
“In the long term, there will definitely be a majority of activities executed by AI-based creatures, robots or algorithms.”
Creatures. Interesting choice of word.
“Things like autonomous driving will definitely be established. There will probably be no drivers on most of the world’s roads.”
“I expect restaurants will be automated. Only the elite restaurants will have a chef! Cafes will be completely automated.”
“Basically, there’ll be no operators of most of existing machines.”
“We need to be ready for this. Ready to reallocate our resources to other stuff where the person is important.”
“We need to be creative. We need to develop these creativity skills because these will be in demand in this future era. The ability to be creative, to think, to analyse and come up with new ideas, to invent."
"Most importantly, we need to be ready to constantly learn.”
댓글