In mid-2017, XAIN won the 1st Porsche Innovation Contest against 120 international startups. This resulted in a partnership between XAIN and Porsche. The initial effort of this partnership was to be first in bringing blockchain functionality into a commercial vehicle, a Porsche Panamera, to explore business cases around this innovative technology. Outcomes of this pilot can be seen in this short video.
As a sports car brand, Porsche places great importance on the experience that its customers get when driving a Porsche vehicle. Yet steering a fast, reliable, and responsive sports car may seem at odds with the current trend in autonomous driving. The owner of a Carrera 4 GTS, for example, may not appreciate when he merely sits in the “driver” seat and the car’s AI agent is in complete control and — perhaps adding insult to injury — when that agent drives the GTS with the seeming perfection of an experienced Formula 3 driver!
Nonetheless, Porsche is very actively pursuing AI as a technology of strategic importance to its brand, for example with its AI Mondays initiative. So how does this apparent tension between an active and exciting driving experience and the use of autonomous AI get resolved? For Porsche, AI is not seen as a means of fully replacing the driver, but of enhancing even further the nature and value of driving a Porsche, as experienced by its customers. AI can achieve this at various levels, let us mention some examples:
- adjusting speed as a function of upcoming traffic a country road,
- predicting maintenance needs in time to minimise service disruption and reduce costs,
- alerting the driver of unexpected road hazards, or
- dynamically adjusting seat and steering wheel positions as a function of who is driving and under which conditions.
Indeed, once more and more cars will have more advanced levels of autonomous driving, it will probably be the luxury car and sports car manufacturers that still want to sell cars for those who want to actively drive their vehicles, but who expect to do this with ample assistance from AI products embedded within those cars.
This intended use of AI in vehicles also means that a lot of the data that gets generated by the vehicle will be of a personal nature. For example, knowing the GPS coordinates for where a Carrera was, and for how long, captures a wealth of private and personal information about the individuals who drive this car. Sensor information about weight shifts, eye movement and such may even be considered as personal health data of the driver. Therefore, the processing of such data becomes subject to regulations such as the EU GDPR, as well as laws that are specific to territories. Car manufacturers can address some of these concerns around data privacy by containing the data footprint, for example by not storing personal data. Furthermore, they can reduce the complexity of data processing so that data controllers have less burden in being compliant with GDPR or other data privacy regulations.
Such measures and best practices, however, grew over time as a reaction to such regulation. They also reflect the nature in which such data is conventionally used: data is retrieved and processed as a response to a programmed query, for example in a commercial database system. However, the use of AI in production means that data is increasingly being used differently, as a training source for the learning of AI models that support decision making, for example the decision to adjust the speed of a Carerra in real-time.
It is therefore not immediately clear how this new AI-driven paradigm of data processing is to be interpreted with existing regulation, nor whether regulation specific to AI and its usage will be introduced in the future.
For AI to be effective, it also requires a huge amount of data for training. This poses a real problem for autonomous driving. Regulation prevents us from merging local datasets that contain personal data into large datasets on which effective learning can take place. To work around this, data in these local datasets could be anonymised prior their merge.
But doing this is not realistic: anonymization requires considerable manual effort, does not scale in the number of datasets, and runs the risk of using anonymization techniques that may not offer adequate protection of personal data — at present or in the future. Wanting to rely on anonymization for data streams from thousands or even millions of vehicles seems thus to be a futile endeavour.
Such uncertainties and adoption barriers around AI may well be seen as risks for AI innovations in autonomous driving. The strategic aim of XAIN is therefore to develop and offer AI technology that can eliminate these risks, and can do this in a way that is agnostic to the particular AI development tools and the AI use case. XAIN achieves this through its unique approach to Federated Machine Learning (FedML).
Our FedML machine-learning technology allows learning of AI models to take place at local datasets. Local data will never have to be moved elsewhere. Rather, the models are trained locally on local data. These locally updated models are then communicated to a Coordinator that aggregates them into an updated global model. All participants of this federated learning then receive this updated global model as the basis for their next local training round.
In this vision of autonomous driving, and of AI assisted active driving experience alike, vehicles do not have to share personal data as such but only locally updated AI models. While the latter are conservatively viewed as also being personal data, XAIN aggregates these locally updated models into a global model in such a manner that the global model is no longer private data and so not subject to the GDPR.
That way, car manufacturers and other stakeholders of the digitization of automotive can meet regulatory demands while also obtaining AI models whose accuracy and quality is basically as high as those one would obtain by a non-compliant merging of all local datasets for training.
Our FedML technology is, and will continue to be, made open source — a very important consideration for software production in automotive. And we firmly believe that our two partnerships with Porsche and with Infineon have synergistic effects: Infineon’s powerful AURIXTM microcontrollers may in the future run AI engines that can locally train AI models within XAIN’s FedML infrastructure, without the need of the raw data to leave the vehicle network and storage systems.
We plan to post more news on our AI collaboration with Porsche in future Medium articles, so please watch this space!
Leif-Nissen Lundbæk is the Co-founder and CEO at XAIN AG and leads their cryptographic developments. For more information on the company XAIN and their eXpandable AI Network, you can find them on Twitter or at xain.io.