Posts in category: artificial intelligence

We have designed a way to leverage the cloud, let’s say, as a means to increase the efficiency of the energy grid. Because the consumption produced by the cloud is by estimate dating from 2016, so pretty much outdated, already taking up 2.5 to 3% of energy consumption in Europe.

And it’s the fastest-growing energy consumer.

And I’m not talking fast as in 10% a year, I am talking fast in doubling every five years.

By some estimates, this energy consumption of IT, of the cloud, in 20 years will reach 20%, of energy consumption, which is unsustainable.

But if we play our cards, right, we can make the number be 7-8-9 percent over the next 20 years.

And one of the things we’re working on is basically using a huge consumer, which is a lot more predictable than households, consumers, right, you can turn it on and off, you can know what effect it has if you turn something off now and turn it on later.

And we sent a proposal, a technical proposal for leveraging this flexibility or huge consumer, in order to fix the energy grid to consume less, when you have less, than to consume more when you need more.

Because one of the key problems of the energy theory right now is what is called the ”duck neck problem”, which refers to the difference between the peak of renewable production and the peak of consumption in time over the span of a day.

So basically, you have the peak of production around noon, and you have the peak of consumption around evening. And basically, that means that you get a lot of green energy when it’s useless, and you have none of it when you need it.

And we can use the cloud to flick the two. And if we do that, just for one in 1000 of the energy grid, the economic advantages are huge, just one, not one person, one in 1000.

And by this sort of idea where people who buy cloud computing, get an option, you can buy it now you can stop the processing now, and you pay this fee, or you can leave the processing for when we have renewable energy in the grid. And it will be easy. If you wait, it’s cheap, and you save Earth.

And you also get a certificate saying that our computation was done using bla bla bla bla bla, and you sign it cryptographically.

And that guy or that company can put it on the website and say look, 80% of our processing is done, not in a carbon-neutral way, in a carbon-negative way.

Because if you reduce the burden on the network, you don’t just use how much carbon you emit. But you also eliminate the need to produce carbon to offset that balance by conventional means, because the way that the way this balance is settled right now is by burning gas when you don’t have energy.

So basically, we have a lot of renewable energy, but we don’t have it when we need it.

And we don’t know how to make batteries, but we know how to turn the cloud into a battery, a negative battery.

But for technical purposes, the negative battery is just the positive battery or a different type. – Bogdan Bocșe, CEO @ Knosis.AI

We need to transform the sort of environmental and sustainable development goals, we need to transform them into perceivable immediate actionable sets, and very offensive, very, not offensive, but very aggressive.

All AI solutions, which are trained exclusively with data coming from a single human person, are biased.

So basically, if you take the word of any single person, you are biased.

So basically, all the solutions that are using judgments or labels or supervision of any sort from a single person per case are by definition of the word bias, and we have solutions for this.

And we have a solution that instead of avoiding bias, measures the types of bias that are being experienced. It actively tests the user request to trick the user of having a biased procedure.

It goes that way to explicitly not having a bias to see if it goes the other way: by removing part of the context to see if it was the other way.

And I will give you a very simple example of a prototype that we built. We haven’t deployed it yet for the customer, but maybe we’re lucky this year. We have this solution for augmenting data and measuring the bias because we wanted to use it for measuring consecutive bias.

We don’t have this product yet. We have a lot of other projects. I won’t do this marketing campaign right now, but the idea is like this, we go to a company and go like:

Do you want to know what sort of biases your Executive Board has?

Cool! Give them this link.

And that link gives you a series of very boring administrative tests: look at some resumes, read some e-mails, but it’s booby-trapped.

For instance, if I show you the profile of a person, I will measure whether you click on the pictures of a certain job, more often than the pictures of something else.

Also if you click on the picture from a CV, you are biassed.

If you click, if you express intention of seeing, you are biassed.

My girlfriend always makes fun of me:

You know, you will have so many girls that you will need to interview, you have so many girls that will be around you. And I go like yeah, but my point is to make sure that I’m a good manager. I want to make sure that I do not put into me information that I know will hurt my decision process.

It is my job to protect myself from being a sexist. – Bogdan Bocșe, CEO @ Knosis.AI