AI Regulations, Car Wreckognition, and External Data Copy
The European Union proposes restrictions and regulations on AI systems in a recent legislative framework.
Artificial Intelligence has gained a lot of momentum in past years. This comes as no surprise, considering the immense potential benefits of the technology for society, such as improved medical care or better education. Some use-cases, however, expose underlying risks that could do more harm than good. Above that, the black-box nature of certain algorithms create further concerns and uncertainties. As a result, governments worldwide are moving to regulate AI.
As Europe enters its Digital Decade, it is looking to enable and fund strong development and uptake of AI. Along with it, it also plans to expand its regulatory arm past data privacy, to reach the models themselves. In particular, the European Commission posits that any AI-generated improvements need to adhere to "rules that safeguard the functioning of markets and the public sector, and people's safety and fundamental rights."
Lawmakers within the European Union have recently presented a risk-based proposal for regulating AI systems. In particular, a three-tier risk separation (low, medium, and high) would define the different regulatory requirements. For example, the makers of high risk applications will have to create records of their systems that look at robustness, accuracy, and security. Mainly, low-risk applications will need to follow good conduct around issues such as environmental sustainability, accessibility for disabled persons, and developer diversity.
The toughest regulations require demonstrating proof of safety, training using only high-quality data, and including detailed documentation with AI systems. Additionally, a certain subset of high-risk systems would be completely forbidden as they pose an unacceptable risk to society. This list specifically contains algorithms that manipulate people via subliminal cues, real-time face recognition, as well as social credit scoring style applications such as the ones used in China at the moment.
The proposed law is intended for application to any company selling AI products or services into the European Union. Companies that violate the new set of rules could be subject to fines up to 6% of global annual turnover. The EU plans to set up a new body, called the European Artificial Intelligence Board (EAIB), to support the application of the regulation.
Today, we aim to make Europe world-class in the development of a secure, trustworthy and human-centered Artificial Intelligence, and the use of it.
Margrethe Vestager, Commission EVP
For more information, the EU has provided a FAQs list that provides good insight into their 100-page proposal.
Why it matters
The European Union is the first governing body in the world to propose such extensive and detailed regulations. With this text as well as the promise of an investment as large as 164 B euros in the coming decade, Europe is sending the world a clear signal: it wants to become a global leader in developing cutting-edge, trustworthy AI.
Not everyone is equally excited about these new regulations. While some believe it will hinder innovation, most criticism states that the regulations are not tough enough. More specifically, the proposed law leaves quite a bit of loopholes for high-risk applications such as automatic gender recognition and exceptions in facial recognition.
Despite the criticism, the new legislation shows that AI's adoption and regulation is an important priority for the European Union's Digital Decade. It creates a platform for advancing the conversation about the development and use of AI systems. An interesting aspect that seems absent from the current text is the fairness and discrimination of AI systems, which will no doubt be the subject of spirited debate in the near future.
The discussed legal framework is only one part of Europe's three-pronged approach to instilling trust in AI:
- European legal framework for AI to address fundamental rights and safety risks specific to the AI systems
- EU rules to address liability issues related to new technologies, including AI systems (last quarter 2021-first quarter 2022)
- Revision of sectoral safety legislation (e.g. Machinery Regulation, General Product Safety Directive, second quarter 2021)
Therefore, we are looking forward to the consolidation of this first regulatory framework and the additional upcoming legislature.
Insurance companies around the globe are turning to Computer Vision powered models to calculate the cost of car repairs.
If you get into an accident with an insured car, the insurance company handles the claim with a repair shop to get your car fixed. This normally entails auto body repairmen to work in collaboration with the insurance company's appraisers and claims adjusters. Usually, a thorough overview of the damage allows to sign off on a fixed amount of repairs depending on the condition of the vehicle.
What happens when COVID-19 strikes, stopping real-life visits and endangering this collaboration?
A new Artificial Intelligence tool assesses damage from minor collisions using mobile phone cameras and predicts the incurred repair costs. High adoption shows that insurance companies around the globe are embracing automation.
An app installed on your mobile phone allows you to take pictures of your vehicle after getting into an accident. It will then use those pictures as well as your response to forms to classify the damage of your vehicle. In turn, the system can estimate the repair cost using availability information from local auto body shops. The algorithms used, which are developed by Tractable, are trained on huge amounts of data from previous claims for vehicles of all makes and models.
While the system needs a human adjuster to review the estimation, it is spot approximately 25 percent of the time. As you can imagine, insurance companies have seen incredible improvements in efficiency, consistency, and timeliness.
Why it matters
One group of workers is particularly unhappy with the innovation: auto body repairmen. In the past, insurance companies have wanted to pay less for repairs than the body shop thought necessary, but it was generally possible to find common ground when they had an appraiser on the shop floor. With this new technology, there is almost no room for discussion. In fact, it is difficult to notice some under-the-hood issues such as suspension damage or a frame misalignment. Evidently, it follows that incomplete estimates lead to incomplete repairs, which is bad news for some vehicle-owners.
A photo is worth a thousand words, but it doesn’t come up with the value of the damage.
Mike LeVasseur, Automotive Service Association collision division
Despite the body shop workers' opinion about the technology, it seems that photo-claims are here to stay. The number of claims settled with photos has risen from a mere 15% to a whopping 60% during the pandemic. Above that, the amount of vehicles inspected per day for each appraiser climbs from 3 to 20, and additionally saves them on cars and gas to move around. The greatest advantage of the technology is that it is in no shape or form a replacement technology. The system remains human-in-the-loop, and essentially supports the appraisers heavy workload.
It seems that the AI-assisted estimates are very good at two things:
- Separating totaled cars from minor damage
- Estimating repair cost for minor collisions
Luckily (and unsurprisingly), these cases make up the large majority of incidents. While these systems are incrementally improved upon, auto body shops and insurance companies are finding ways to collaborate using this new technology.
Writers from MIT Sloan insist on the utility of external data for a company's analytics in a recent article.
The best data strategy allows a company to turn data into important insights in all steps of the value chain. The ultimate goal is generally along the lines of reducing costs, optimizing processes, or increasing revenue. For years, the data utilized in that equation has been intrinsic data, meaning it was created or recorded within company walls.
This shines light on an interesting question: if an entity has external relationships and dependencies with other entities, why not use external data to optimize decision-making with regards to those relationship?
You might think this is obvious. However, while most companies track information about their interactions with different entities, they often miss out on an important part of the picture. Too often, companies don't look outside their walls and forget to incorporate external data into their workflows! Think of a wholesale company that sells products to other stores. As you can imagine, previous purchases and interactions between the company and its clients are used to gather insights to optimize future purchases and client satisfaction. In this scenario, that company might be forgetting key external data such as third-party information about their clients, macro-economic trends, or even weather data.
A recent article by MIT Sloan explains Why external data should be part of your data strategy. The use of third-party data is relevant in all sorts of different industries and use-cases. It is, however, especially useful for time-series predictions. In fact, the Vice President of Data Strategy of Hartford Insurance states that "you can't build high quality predictive models with just internal data".
As you can imagine the use of external information is not necessarily novel. A study by Deloitte states that a whopping 92% of data analytics companies have used external data sources to ameliorate their insights generation.
While it seems obvious for a lemonade stand to predict sales using the weather forecast, it's not always as easy to know what data to incorporate. The first reason is the incredibly vast data market. You can find tons of free and curated datasets online on government and non-profit organization websites. However, most use-cases require the use of specific data, such as regulatory, property, weather, and telematics data, which usually comes at a price. It is not uncommon to find analytics companies that have their own data-hunting team whose job is to scour for useful data sources outside of the company for given use-cases. Data is becoming a key business asset for most companies, and they are stacking up the best procurement teams to future-proof their analytics.
Why it matters
92% of data analytics companies have used external data sources to ameliorate their insights generation.
The incorporation of external data is an essential part of a company's analytics as it helps them gain strategic insights from outside their walls. Key for knowing your customers better and adding a real-world context to internal decision-making, third-party information is the key to a successful data analytics function in any company.
Don't forget that third-party data is only useful when you have a system to use it with. A strategy including this data is successful if and only if it is part of a larger digital strategy.
Another key consideration for firms trying to build up their data maturity is that fitting an external data component into an existing system is time-consuming. With its incredible added value comes the responsibility of curating that information, making sure that they are not in violation of privacy protection regulations, and that they are unbiased and accurate.
At Visium, we are extremely keen on building Machine Learning systems that incorporate external data. With vast experience in handling the processing and joining of the third-party information into your existing system, we deliver high-end, ready-to-use systems. If you are interested in stepping up your company's use of external data and build something great with us, contact us here.