Chair @ Deel Lab | Building Legal AI l Driving Global Work Policy

I lead the Conflict Analytics Lab (CAL), a frontier AI research lab focused on law and dispute resolution. Our work examines how advanced AI—especially reasoning systems—can support access to justice and improve legal decision-making. A central part of this research explores how lawyers and AI can co-design reliable legal technologies.

I also founded OpenJustice, an open-source platform spun out of the Conflict Analytics Lab. Using a no-code method, lawyers encode structured logic, constraints, and experiential knowledge directly into AI systems. The aim is to reduce hallucination and sycophancy by adding reasoning layers—not more data—thereby improving reliability, alignment, and performance in high-stakes legal tasks.

Previously, I served as Chief of Policy and Head of the Deel Lab at Deel, where I led the development of several applied AI projects, including the Deel AI Classifier. I am an Associate Professor at Queen’s University Faculty of Law, with visiting faculty appointments at Cornell University and Paris Dauphine University, and past academic affiliations with ENA and Harvard University.

Earlier in my career, I served as a Cabinet Member at the Court of Justice of the European Union, advised the European Commission on financial aid programs in Eastern Europe, and worked as a conflict resolution consultant at IRENE–ESSEC. I also clerked for the French Administrative Supreme Court (Conseil d’État).

💯  agree but I don’t think Canada has an AI talent problem.

We have a follow-through problem.

Yes, students should stay in Canada. Our research system is strong, and there are many ways to build excellent labs and teams here - thanks to Natural Sciences and Engineering Research Council of Canada (NSERC) Mitacs  Social Sciences and Humanities Research Council of Canada (SSHRC) and many others

But the real is issuse: research happens in Canada → value creation happens elsewhere and it gets funded well.

US capital comes in, turns research into global industries — and often relocates the companies.

The impact is real, which is good.

But too often, the long-term benefits aren’t captured in Canada.
That’s not sustainable.

If we want people to stay, we need to make it easier to build, deploy, and scale research in Canada — not just publish it.


Minister Evan Solomon 
https://lnkd.in/ejqFb26E

Samuel Dahan

Law Professor l Frontier Research for Open-source Legal AI | Co-founder @Openjustice & Conflict Analytics Lab l AIPP @Mila

Legal AI keeps getting better at sounding right.
 That hasn’t made it more reliable.
The problem isn’t model size—or adding more text, as if textbooks and cases fully capture lawyering.
 It’s the kind of legal knowledge we put into these systems.
There are new ways to instill legal knowledge and reasoning into language models—ways that reflect how lawyers actually work.
Dr. Megan Ma Prof. Xiaodan Zhu Chu-Fei Luo and I  tried to unpack this in a recent piece for the The National Law Review: why the next phase of legal AI depends less on prompting, and more on lawyers encoding reasoning from practice.

Thanks to Oliver Roberts 
https://lnkd.in/g3z5YCv9

The Globe and Mail on Canada’s “legal data desert” — out today

 An investigation by Robyn Doolittle, published this morning, looks at a quiet but serious issue in Canada’s justice system:

We don’t have a complete, public record of court decisions.
Here’s why that matters.

In Canada:
- judges decide what gets published
- courts voluntarily send decisions to CanLII
- some decisions go to commercial databases
- many never become easily accessible at all

There is no centralized, public system that guarantees full access.

CanLII does important, open-access work — and deserves credit for it.

But public legal data is still fragmented and not open for bulk access.

That creates two downstream problems.

First: AI becomes winner-takes-all.
Without bulk legal data, only a few actors can build serious legal AI tools. Less competition → very expensive products → limited access for legal aid orgs and solo practitioners — often the people who need these tools most.

Second (and deeper): you can’t evaluate justice without the full picture.
If we can’t see all decisions, we can’t tell whether similar cases are decided consistently.

Inconsistency doesn’t disappear — it just becomes invisible.

Other countries have moved differently (France, the UK, the U.S.).
- with Free Law Project Michael Lissner in the US; #Judilibre Édouard Rottier ; #CambridgeLegalCorpus Felix Steffek

Canada is increasingly an outlier 

This isn’t just a tech issue.

It’s about transparency, accountability, and the rule of law in a digital age.

This raises a bigger question:
Can we really talk about access to justice if we can’t fully see how justice is applied?

https://lnkd.in/e6xU8nb3

There’s a lot of enthusiasm right now around agentic systems and “#vibecoded” AI - even in LAW

And rightly so — they’re fast, creative, and often impressive.

But this image (below) captures something lawyers should pause on.
In legal AI, the real risk isn’t messy code.

It’s deploying uncertain legal outputs at scale.

Agentic systems are, at heart, a form of output deployment.
 If the underlying legal reasoning is inaccurate or only partially correct, agents don’t fix that.

They replicate it, route it, and embed it into workflows — where mistakes become harder to detect, harder to explain, and harder to unwind.

That dynamic feels uncomfortably familiar.

A bit like 2008:
opaque products
layered delegation
errors passed quietly from system to system
until no one can clearly trace where the problem started
Agents are powerful.

But in high-stakes domains like law, speed is not the binding constraint.
Substance is.

We’re still seeing error rates north of ~50–60% in many legal RAG setups.
 Scaling that through agents doesn’t create leverage.
It creates systemic risk.

For law firms and legal institutions, the priority should be:
getting the legal content right
making reasoning inspectable and traceable
then deploying agents on top of something lawyers can actually stand behind

Agents will matter. A lot.
But first we need to scale correctness, not confidence.
—
Image and original framing inspired by a post from Andreas Horn.
his newsletter: 𝗻𝗲𝘄𝘀𝗹𝗲𝘁𝘁𝗲𝗿: https://lnkd.in/dbf74Y9E


#LegalAI #AgenticAI #ResponsibleAI
#vibeCoding

Another reason for Canada to build its own deep-tech investment system

With Y Combinator stepping back from Canada, it’s a reminder that foreign capital doesn’t always align with where value is built long-term.

The flip side is encouraging. As #MarkCarney noted at World Economic Forum, Canada has capital, talent, strong pension funds, and a very competitive environment for investment.
 
The opportunity: public capital, pensions, and private investors (both Canadian and Foreign) working together to anchor pre-seed and seed funding for deep tech and research-driven companies in Canada

PhD & Postdoc openings | NLP · LLMs · Machine Learning · AI for Law

We’re hiring fully funded PhD students and a Postdoctoral Fellow at Queen’s University, working at the intersection of NLP, LLMs, and high-stakes decision systems.

The lab is jointly led by Xiaodan Zhu (leading NLP scholar, full member of the Vector Institute) and myself, bridging core NLP/ML with real institutional deployment in law.

Why this is interesting (for CS folks):
Legal AI fails less because of model size than because knowledge, reasoning, and constraints are not represented explicitly. Our research asks how LLMs can be augmented with structured reasoning, domain logic, and evaluation frameworks that actually hold up in adversarial, high-stakes settings.

A central research infrastructure is https://lnkd.in/gjBNKVTu open-source platform for instilling legal knowledge and reasoning into LLM-based systems. It is used as a shared experimental environment, not a product.

OpenJustice is supported by a 40+ partner network, enabling empirical work beyond benchmarks:
Industry & practice: Deel , Osler, Hoskin & Harcourt LLP  Borden Ladner Gervais LLP (BLG), Dentons, Edilex
Access-to-Justice research: Access to Justice Lab at Harvard Law School Lab, Stanford Law School @Legal Innovation through Frontier Technology
AI & CS collaborators across NLP, ML, HCI, and law

Research directions
-LLM-augmented reasoning systems
-Legal language understanding & text analytics
-Trustworthy / responsible AI in high-risk domains
-Evaluation of AI systems embedded in real workflows

PhD (fully funded)
Start: Sept 2026 or Jan 2027
Preferred deadline: Jan 31, 2026 (rolling)
Postdoc
Start: from Jan 2026
Duration: 1–2 years
📩 Contact: david.liang@queensu.ca

📄 Apply: Queen’s ECE graduate admissions

If you work in NLP or ML and care about how models behave when the cost of being wrong is real, this is the kind of environment we’re building.
Please feel free to share with students or colleagues.

Intellectual environment
Our extended research network includes scholars, fellows, and collaborators who come from or are affiliated with leading AI research institutions and communities, including Mila - Quebec Artificial Intelligence Institute  , Hugging Face (particularly evaluation and open-model teams), @Allen Institute for AI, Stanford Human-Centered AI Institute, and the Vector Institute NLP Community.
Students and postdocs in this program will work in an environment that regularly engages with leading scholars in legal AI, NLP, and applied ML, including collaborators and advisors such as:
Maxime Cohen
Jonathan Touboul
Dr. Megan Ma
Aurelia Tamo-Larrieux
D. James Greiner
Mandy Mobley Li
Yuri Levin
Mikhail Nediak
Derek Nowrouzezahrai
This means close exposure to cutting-edge work in NLP/LLMs, empirical evaluation, decision systems, and real institutional deployment—not just siloed technical development.

Joel Niklaus

Stephen J. R. Smith Faculty of Engineering and Applied Science at Queen's University

View company: Stephen J. R. Smith Faculty of Engineering and Applied Science at Queen's University

Fully funded Master's, PhD, and Postdoc positions are now available. Specific research projects and a list of supervisors are here: https://lnkd.in/gDidh9jH 

Please share widely to help us find great candidates for these positions!

No alternative text description for this image

taplio

Samuel Dahan's Linkedin Analytics

Samuel Dahan

Performance ScoreSamuel Dahan's LinkedIn Performance

What is Samuel talking about?

Samuel Dahan's Best Posts (last 30 days)

Ready to grow your LinkedIn presence?

taplio

.css-6ytdur{background-image:linear-gradient(284deg, rgba(39, 43, 72, 1) 0%, rgba(33, 47, 183, 1) 100%);color:transparent;-webkit-background-clip:text;background-clip:text;}Samuel Dahan's Linkedin Analytics

Samuel Dahan

Performance Score.css-goub6s{font-size:var(--chakra-fontSizes-md);font-weight:500;color:var(--chakra-colors-gray-600);display:block;margin-top:var(--chakra-space-1);}Samuel Dahan's LinkedIn Performance

What is Samuel talking about?

Samuel Dahan's Best Posts .css-1emoo84{font-size:var(--chakra-fontSizes-md);font-weight:400;color:var(--chakra-colors-gray-600);}(last 30 days)

Ready to grow your LinkedIn presence?

Samuel Dahan's Linkedin Analytics

Performance ScoreSamuel Dahan's LinkedIn Performance

Samuel Dahan's Best Posts (last 30 days)