I attended EGG AI conference yesterday – an AI conference which put the spotlight on real-world use cases and hot topics like machine learning interpretability, bias, and fairness. There were talks from data analysts, engineers and scientists at top companies who are using AI in their products, operations. 

It was the first time I’d attended a conference like this and although the topics and crowd were different from what I’m used to at UX and UI conferences there were lots of similarities. What really stood out to me was the passion everyone had about data. In the same way designers are passionate about good design which empowers the user, the people here were passionate about accurate and ethical datasets that can improve peoples’ quality of life. I left the conference with a new passion for data!

Below are some notes I made throughout the day:

Reducing Pollution in London

Kim Nisson, CEO at Pivigo gave a fascinating talk about a project investigating how they could use data and machine learning tackle air pollution in London.

Air pollution in London is a big problem. I can’t remember the figures, but it’s the cause of many deaths and health issues each year. The Mayor of London wants to tackle London’s pollution with lots of local interventions street by street. There’s a big budget to fund this but London’s a big place – how do they know where to spend the money? Where do they start?

Ideally, they would be able to measure the air quality of each area of London but there’s a problem – London only has 100 sparsely distributed air sensors. Not nearly enough. It does however have over 700 traffic cameras. Could they use traffic cameras to measure air pollution? Could they combine weather data to further enrich their model to estimate pollution levels?

The aim of the Pivigo’s project was to improve London’s ability to measure air quality as this is sparse so the government knows where to apply interventions. They used deep learning to recognise vehicle types on traffic cameras. They could then map traffic intensity with pollution in areas with air quality sensors and found there was a reasonably good correlation. After this, they developed an Air Quality Prediction Model which predicted air quality based on traffic and weather data.

This tool was used to understand the most polluting vehicles and worst affected areas which will feed into a strategy for lowering air pollution in London and improve the health of the population. This can be done using proxy data, without the need for putting up many expensive air quality sensors.

This is a great example of how data helps ensure investments are made in the best way possible.There were also similar projects to predict where in London to put Santander bikes and where to issue flood warnings.

The Good City Life Project

Luca Maria Aiello from Nokia Bell Labs gave an interesting talk about how his team at Goodcitylife.org is improving life for city-dwellers.

The corporate smart-city rhetoric is about efficiency, predictability, and security. “You’ll get to work on time, no queue when you go shopping, and you are safe because of CCTV cameras around you”. All these things make a city acceptable, but they don’t make a city great. 

Efficient doesn’t necessary equal quality of life. You must consider psychological perceptions. To improve the experience of living in a city, you must understand the psychological perfections of living in a city. 

The Good City Life project asked ‘Can we measure the experience of living in a city?’ Turns out they could…

The first area of focus was beauty – how attractive parts of the city were. Urbangems.org allowed people to select the most beautiful photos from choice of two. The data collected here helped assess how beautiful a city was. This involved a lot of work from people manually selecting the photo they found most beautiful which wasn’t scalable. It was made more scalable by training an AI to learn the assessment of beauty so it could rate photos.

This data was then used in Happy Maps – an app that recommended users directions through the most beautiful, pleasant locations. Even if it took slightly longer, it would seem shorter as it was a nicer walk. 

The feedback on Happy Maps was good but people felt there was something missing – smell; ie. I may walk down a beautiful alley but it smelled of smog. Smell impacts our experience of space as much as sight. Sending people around cities manually measuring smells was not scalable. So they created a taxonomy categorising different smells and each smell’s pleasantness. They then posted photos to social media, asking users to add smell tags to photos taken around different parts of the city. Smelly Maps was then launched.

The same was then done with sound and culture, where they mapped the cultures around different parts of the world. This data was useful to predicting future rent prices in different areas. 

After that, they mapped food consumption: A map of types of food consumed by people in different parts of London to predict where people would likely fall ill. For example, where people consume the most fat or sugar. 

These maps weren’t only nice, but they were useful and delivered value.

Misconceptions about AI

Over the last few years we’ve seen an explosion of complexity in our data science models but complexity is not intelligence. It’s not AI, it’s machine learning models.

  • Corporations have a false sense of progression in AI. From a business perspective they are celebrating AI as a new hype but the reality is that we’ve just got more complex machine learning models.
  • The public see data science being a way for corporations to make money. Whilst that’s true it’s not the only benefit. In the future, data science will improve people’s lives. Data science will improve areas such as healthcare but the public don’t realise this yet. We need to show people how data can improve people’s lives.
  • Data scientists want to do cool things. They should understand the business needs and the business application of what they do. A lot of companies struggle to get things into production and for this we need technical and business application to combine. People getting into the field think ‘I’m going to go into a dark room and get into crazy cool stuff’ but in reality they need to talk to people and promote what they do. Understand stakeholder needs.
  • Data scientists should talk to stakeholders to understand the problem being solved. They should communicate the problems their data models solve differently depending who they’re talking to so it can be tailored to who they’re talking to. The right pitch for the right audience. 

Ethics

  • AI is a tool and it doesn’t change companies responsibility to be ethical. But it can be complex and not clear how some decisions or recommendations are made. There may be unfair biases in the system without you realising. You should be able to explain to clients why a system is behaving the way it is. That’s an area organisations are struggling with.
  • Discrimination: Advertisers claim that they are not interested in who the person is but their traits. ie. they care about what you watch, read etc. But what does it mean if you are being discriminated against based on your traits? Does this create social groups we haven’t defined yet? Users can’t create a claim against this kind of discrimination as it’s not about them, it’s about what they do. 
  • Every data set is biased so it’s important to do due diligence to find out where these biases may be and not make software which May push one way or another. Be aware of impacts and ensure your models align with what you’re trying to do.
  • Fiction is becoming reality. It now doesn’t seem too far fetched. Are we ready for this application of AI? Do we want it?
  • If AI is the new gold rush, you (data analyst/engineer/scientist) are the pick axe vendor.
  • Drive by numbers can make us ignorant.
  • In a general sense, AI is powering so much of what’s around us. It’s important to regulate appropriately. AI ethics is the gap between what the technology enables us to do and what the law and society believes is the right thing to do.
  • Morals change over time and contexts shift. How you deal with these things is really important.

Tidbits

  • The challenge is that data is everywhere but people don’t use it. Making data meaningful is important to get people to start using data.
  • Align data with end-user. Start with the persona – what they trying to do. Then tie the data services back into how it helps the end user.
  • You won’t replace people and operations with AI but you will be able to augment people with data to improve efficiency. Efficiency means less time spent on incident management and more time spent on problem solving. It shifts the effort to the more fulfilling parts. There is concern that efficiency means ppl will be out of jobs but it means that people will be doing higher value tasks.
  • AI is a tool that appears to be intelligent. We’re building seemingly intelligent tools. It’s just the next gen of tools. It’s never going to replace the Mark 1 Human (hopefully).
  • Data is like Lego – the building blocks of a business. One block on its own isn’t much, but build them together and it can be powerful. The possibilities are endless.
  • You do not want to be a data driven company. You want to be a data embracing company. Our companies exist for a purpose and we want the data to underpin the business strategy to help the company with its purpose.
  • Could a CEO be an AI?