AI Expert Witnesses

Hiring an AI Expert Witness

One of the biggest challenges in legal cases involving AI is that AI systems are hard for humans to understand, as they are not ‘programmed’ in the traditional sense. Traditional computer programs function as a series of instructions, like a recipe, while AI systems write their own instructions. This means that it can be hard for fact-finders to follow how a system adapts.

In such cases, it is essential to find an expert in AI who not only understands the AI system fully, but is able to explain the key points to a lay audience.

AI expert witnesses typically come from fields such as software engineering or robotics, and may specialise in questions on niche areas of machine learning such as neural networks, or natural language processing (NLP).

AI Expert Witnesses at Fast Data Science

At Fast Data Science we have a number of data scientists in our team, and our main focus is natural language processing (NLP). The manager, Thomas Wood, studied a Masters in 2008 at Cambridge University in an area of NLP, Computer Speech, Text and Internet Technology, and conducted his research project on pleonastic pronouns using unsupervised learning. Since completing his postgraduate studies he has worked exclusively in data science, maintaining a constant focus on NLP, although he has occasionally worked in computer vision and other areas of data science, including a stint consulting for Tesco, predicting customer purchases. Thomas Wood founded Fast Data Science Ltd in 2018 to deliver data science consultancy focussing on natural language processing problems in large organisations that deal with lots of text data, such as healthcare, pharma, insurance and legal. In 2020 Fast Data Science diversified out to offer services in AI due diligence, and AI expert witnesses.

AI Expert Witnesses in the UK

In England and Wales, the Crown Prosecution Service and the Civil Justice Council have published guidance on the use of expert witnesses in civil and criminal cases. Expert witnesses can be contrasted with fact witnesses. The evidence given by an expert witness must be tendered as opinion rather than fact.

“Expert evidence is admissible to furnish the court with information which is likely to be outside the experience and the knowledge of a judge or jury.” Ministry of Justice (England and Wales), Criminal Practice Direction V Evidence 19A Expert Evidence

For further information about expert witnesses, please refer to the following resources:

• Crown Prosecution Service guidance to the use of expert evidence in criminal cases.
• Civil Justice Council guidance for expert evidence in civil claims.

What is expert evidence?

Expert evidence is simply the opinion of the expert. An expert witness assists a court in reaching a decision by providing independent technical analysis in their field of expertise, and an opinion on the issue at hand.

Expert evidence is admissible in Court when there are open questions that require the input of an expert. The Court has the power to permit or exclude expert evidence. In some large complex cases, each party may instruct their own expert witness, and a Court may allow the appointment of more than one expert.

Our AI expertise

Our data scientists have expertise in many areas of AI, in particular natural language processing, and including:

• Simple vanilla models, such as Bag of words, tf*idf, cosine similarity. These often serve to provide a baseline performance before progressing to more advanced models.
• Slightly more sophisticated models, taking word order into account, such as NLP pipelines, lemmatisation, parsers, chunkers.
• Cutting-edge models such as deep neural networks
  • convolutional neural networks (CNNs; text as well as images)
  • RNN, LSTM
  • BERT, ELMO
  • Seq2seq, word2vec, doc2vec
  • see a live demo of a CNN for author identification
• Clustering and unsupervised techniques
  • Latent Dirichlet Allocation - LDA is useful for extracting topics from a set of unstructured documents, for example, legal documents, survey responses, factory error reports, etc, where there is just an abundance of documents but no accompanying structured data or labels which could make the NLP task easier.
• Search engines and search term recommendation systems
• Google Natural Language, AWS, Microsoft Azure

*Topic detection is a technique used by NLP data scientists to explore and discover common themes in a set of unstructured documents such as factory error reports.