Sign In

Phil Lee

Managing Director, Digiphile - Data advice that is Simple. Strategic. Actionable.

Phil Lee is the Managing Director at Digiphile Services Limited, as indicated by his LinkedIn profile.1 He is a commercial lawyer with expertise in legal matters related to data protection and interactive entertainment.1

Professional Background

Phil Lee has established himself as an accomplished professional in the legal field, particularly in areas intersecting technology and law. His expertise includes:

  • Data Protection: He has significant experience in legal matters concerning data protection.1
  • Interactive Entertainment: Lee is knowledgeable about legal aspects of the interactive entertainment industry.1

Additional Roles and Recognitions

Beyond his position at Digiphile Services Limited, Phil Lee holds several other notable roles:

  • Director of Pembroke Privacy2
  • IAPP Country Leader for Ireland2
  • Westin Emeritus Fellow2
  • Fellow of Information Privacy (FIP)2

Professional Network

Phil Lee's LinkedIn profile (username: phillee77) serves as a platform for his professional connections and activities. He occasionally shares insights and updates related to his field of expertise.2

Phil Lee's combination of legal expertise, focus on data protection, and involvement in the interactive entertainment sector positions him as a valuable professional in the intersection of law and technology.

Highlights

Sep 5 · linkedin

In practice, the most common use of AI we encounter at Digiphile concerns AI integrations - i.e. where one company integrates another's AI, for example a company integrating a third party's generative AI into their products and services. Yet, despite this, the treatment of AI integrations under the #AIAct is surprisingly vague and ambiguous, and little appears - to date - to have been written about the treatment of AI integrations under the AI Act.

To give just a few examples of the unanswered questions that exist:

❓ If you integrate a third party's AI into your products, does that in turn also make your products as a whole an "AI system"?

❓ If so, does that make you a "provider" or "deployer" under the Act?

❓ If you integrate a third party AI model which in fact executes on the third party vendor's servers (e.g. because you integrate via a remote API call), are you actually integrating an AI system, rather than an AI model?

❓ And how should responsibility for the integrated AI be apportioned between the company integrating the AI and its upstream AI vendor?

To some extent, you could say that this all matters only if the integrated AI, or resultant product with integrated AI, falls within one of the AI Act's regulated risk categorisations, but understanding the relationships between vendors, customers, and those customer's customers is important for all sorts of reasons - not least for commercial relationships, contractual warranties and apportionment of risk. Yes, there is some language in Art 25 of the AIAct, but it's far from clear or helpful.

I've been brainstorming these issues with a couple of fellow AI nerds (thanks in particular to Victoria Hordern and Dr W Kuan Hon) and sketched out below how I think these integrations work, covering both integrations of AI models and AI systems and across both B2C, B2B, B2B2C and B2B2B relationships.

Comments from the LinkedIn hive mind are, of course, welcome - we all have to puzzle this out together.

Sep 3 · linkedin

Why is algorithmic design so important to protect against discrimination? Here's a simple example.

Let's imagine you want to implement fair hiring processes for your company, ensuring you hire equally across race and gender. For the sake of simplicity, also imagine that the candidate population comprises just two races (blues and greens) and two genders (men and women). You train an algorithm to help you recruit candidates in a fair manner.

However, a "solution" your algorithm determines to satisfy your objective is to hire only "blue men" and "green women". In doing so, you would hire 50% blues, 50% greens and 50% men and 50% women. You would also wholly (and unintentionally) discriminate against "green men" and "blue women".

The problem here is that the algorithm was trained only to ensure fairness in the top-level attributes of race and gender - not the subgroups that can be created by combining these attributes. The solution it landed on was an accurate one to solve the objective - but also self-evidently discriminatory and unfair.

Although a simple example, it highlights how, without careful algorithmic design, the potential for unintentional discrimination is significant, even in "accurate" algorithms - and especially when today's algorithms collect significantly more data than the limited number of attributes used in this example.

(Credit here to "The Ethical Algorithm" by Michael Kearns and Aaron Roth, where I've taken and adapted an example they give. Thanks also to Ruth Boardman who recommended the book to me in the first place.)

Related Questions

What are Phil Lee's main responsibilities at Digiphile Services Limited?
How did Phil Lee transition from a commercial lawyer to a managing director?
What are some notable projects Phil Lee has worked on at Digiphile Services Limited?
How does Phil Lee's experience in data protection and interactive entertainment benefit Digiphile Services Limited?
What other roles or positions has Phil Lee held in the past?
Phil Lee
Phil Lee, photo 1
Phil Lee, photo 2
Add to my network

Location

London, England, United Kingdom