5 Qs with Montel AI
As power producers seem set for greater levels of investment as part of the global green recovery from the COVID-19 pandemic, Sigfred Sørensen, Software Developer for Montel AI takes a look into one technology underpinning progress in the sector and how new technologies are reducing costs for these generators.
This blogpost is authored by a member of Montel's content marketing team.
How did Montel AI begin?
Back in 2016, the business was born out of a very simple question: with all the data that Montel has access to, why were we not doing any data analysis?
Thanks for that particular insight must go to Olav Edland, an external consultant that asked the right questions at the right time. As an open question to all departments of Montel, it became obvious that there was clearly some potential for something to happen.
In what has now become the AI team, we already had the know-how, the drive to do it and access to data, all that was needed was that initial spark.
The first year was mostly used to learn the technology and to create our own tools for working with it. After some time spent learning, we decided that price would be the first challenge we would try to seriously tackle. We got some excellent help from Norsk Regnesentral (Norwegian computing centre) to understand how to build classical forecasting models and began carving our own path into model building and forecasting.
AI is often seen as a futuristic technology, who can make use of it right now?
Everyone! There are plenty of ways to use the technology, but the important part is having access to the right knowledge and data to use it properly.
I have to say here that AI cannot solve everything; so it’s important to identify what kinds of problems you can solve with it. For us, the technology makes a lot of sense in scenarios where we can see room for steady improvements – in fact the harder the problem is, the better AI is likely to fit the task. On the other hand, if you have a problem that must be solved exactly, AI is probably not for you.
A defining trait in intelligence is the ability to fail and use it to learn and improve. This feature of failure also applies to AI. What matters is really the economy of the problem, or ‘is the problem solved better than any human can’ for example in AI image recognition; are there less accidents or deaths when using an AI driver; is the forecast more precise on average or on specific events than existing models; or can it play chess better than the current best human-created computer algorithms. These kinds of scenarios are never 100% solved, but improvement can come continually over time.
For ANI (Artificial Narrow Intelligence) where you try to learn and improve on one specific task, this is likely to be a slowly evolving process, where we will see some new types of products and gradual improvements year over year. For us in Montel AI however, AGI (Artificial General Intelligence) is the true future technology, one in which you strive for the ability to understand more than one specific task and maybe even achieve self-awareness. AGI is a totally different beast and therefore not likely to be a slow change when and if it arrives.
What makes Montel AI so useful?
I think it’s the ability to create nearly any model for any hard or inexact task. We are not all energy experts within the problems we solve, and this is actually a considerable advantage. We are technology and model experts and will use external industry expertise to guide us in what way our models should behave.
Much like with modern AI engines for chess, the creators do not need to be good or even average players to create the strongest chess computer algorithm in the world, they just need evaluation and guidance.
A defining trait in intelligence is the ability to fail and use it to learn and improve... What matters is really the economy of the problem, or ‘is the problem solved better than any human can?"
Every emerging technology has challenges to overcome, what are they for AI?
Now this is a very difficult question. Everything is a challenge when it comes to AI model development, nothing is set, and the technology landscape is shifting rapidly under our feet. There is no guidance or examples on how to do anything, everything is up to you and your team.
With AI you have every conceivable problem. You can have a lot of data but can never get enough computing power, meaning you need to economise for speed, throughput, and power to improve. Or sometimes you have small amounts of data, and you must try to economise for efficiency and accuracy at the cost of power and speed. Every problem poses its own challenges.
More generally, there are some public concerns with trust issues around AI and the vast amount of computing and power needed to solve some problems. These are valid concerns, but from my point of view, there are other issues and challenges that might be more concerning.
For many tasks, AI is ultimately dependant on access to and the quality of data. This has given rise to a new term: data access strategies. This can be the gathering of data through your social network usage, or your car might be gathering data on how you drive for example. Data gathering is everywhere already, but users tend to have little knowledge of this activity.
Data is gathered and created with the purpose of improving the algorithms in a self-reinforcing way; more users lead to more data, which can then be used to make a better algorithm. With a better algorithm, you get even more users, creating a never-ending cycle. Of course, this sounds great - who does not want a better app or service! – but the downside is that it becomes it very difficult or even impossible to compete with those that are established in the field. As such businesses then run the risk of success not by being customer friendly, but by being first or largest. The business that owns the data will forever be king.
This alongside increased public awareness of the gathering, hoarding, and brokering of user data could become a major issue in the near future. Everyone wants better efficiency and more trustable algorithms, but there is no business incentive to be open and share data for the common good at this point in time.
What does the future look like for AI?
For Montel AI we do not really know our true path yet. Our mindset is focused on creating a good machine learning environment. For now, we have had successes related to the Montel Group’s area within the energy sector, but who knows what we will be doing in 10 years. We do not feel that we are bound to any markets in particular.
For AI in general we are especially excited to see what new types of products that will emerge. Maybe within the next 10 or 50 years, the near future could even include true artificial general intelligence. And as for where we stand? Well, some on the team are a bit wary of the impact that technology might have on society, but others eagerly await our new robot overlords!