Building Trust and Transparency in AI: Insights from Neo4j's CEO

The Importance of Transparency and Trust

Emil Eifrem, CEO and founder of Neo4j, believes that transparency is fundamental to building trust - both in business and in AI systems. As he puts it:

"The absolute flip side of transparency is trust. How do you truly build trust? It's through transparency. We've seen that time and time again in almost all societal processes."

This philosophy has guided Neo4j's approach as an open source graph database company. By making their technology transparent and accessible, they've been able to build trust with developers and customers over time.

Neo4j's Open Source Origins

Neo4j's origins trace back to the late 2000s, when Eifrem and his co-founders were building enterprise content management systems in Sweden. They kept running into limitations with traditional relational databases when trying to model complex, interconnected data.

As Eifrem recalls:

"We realized many of the problems we had, if only we could think of the data in terms of networks - nodes connected to other nodes through relationships. Hallelujah, that solved actually all these problems!"

This insight led them to create a new type of database optimized for graph data structures. They decided to open source the technology, both as a distribution mechanism and because it aligned with their values:

"When we sat down as an early team, it was very obvious which type of company we wanted to build," says Eifrem. "It just felt like a much healthier and better way of building a good company."

Open sourcing allowed Neo4j to build trust and gain adoption among developers. It also enabled a global community to form around the technology.

Evolving the Company Mission and Values

As Neo4j has grown over the years, Eifrem has had to navigate how to stay true to the company's core mission and values while also evolving as an organization.

He emphasizes the importance of regularly refreshing and re-articulating company values:

"We've done a lot of core values refresh. We've tried to do that in an authentic bottom-up way, not entirely by the way - it's not an entirely democratic process, but with a lot of buy-in."

This process helps ensure values remain relevant and descriptive of how the company actually operates. It also gives employees a chance to shape the culture.

At the same time, Eifrem acknowledges the challenge of avoiding mission drift:

"I have that insecurity right now. I wouldn't say that [our values] constantly change - we've changed them probably around every four or five years. But I've not figured out a good recipe for identifying that small angular velocity where you start seeing the drift."

He stresses the importance of intellectual honesty and regularly examining whether the company is staying true to its authentic self.

Navigating the AI Revolution

The recent explosion of generative AI and large language models has created both challenges and opportunities for Neo4j.

Eifrem describes the tension between short-term operational pressures and the need to invest in AI innovation:

"When [the AI boom] happened, we had launched a graph data science product targeted at data scientists. So we kind of dipped our toes into the world of data science and AI a little bit. But when ChatGPT happened, it was really that moment for most of the world."

"These were very tough years. This is just as the industry got off the 10-12 year bull run. How many of your CEO friends said 'I'm in wartime mode'? Probably every single one of them."

Despite the pressure to focus on short-term results, Eifrem recognized AI could be a massive opportunity for Neo4j. He made the difficult decision to carve out resources for AI innovation:

"We pulled away just a tiger team, a small team - a handful of people. It was five people. They had to be more than 80% of their time focused on this. And then I spent a huge amount of time with them."

This bet has paid off, positioning Neo4j to play a key role in the emerging AI ecosystem.

The Critical Role of Retrieval in AI Systems

One of the most important and often overlooked aspects of modern AI systems is retrieval - the process of finding and surfacing relevant information to augment AI models.

As Eifrem explains:

"The LLMs are only trained on data up until the time when you stop training them. People don't realize this - they are complete amnesiacs. They don't remember anything that you told them a second ago. Every question you ask is a brand new question as far as it's concerned."

This is why retrieval augmented generation (RAG) has become so critical for enterprise AI applications. RAG allows AI systems to access up-to-date information from company knowledge bases and databases.

Eifrem draws a parallel to the early days of web search:

"There is a massive case study of doing retrieval - the biggest one that humanity has ever done, which is called the web. There were actually search engines before Google, like Lycos and Excite and AltaVista. But no matter what I searched for, I got like a million hits, but none of them were relevant."

"The reason that Google today is one of the most valuable companies on the planet is that they solved that problem. They solved that problem by identifying the top 10 blue links - the 10 most relevant documents for you."

Google achieved this through graph algorithms like PageRank. Similarly, graph-based approaches are now being applied to retrieval for AI:

"That is exactly what we then do in this world of AI, which is called RAG or specifically graph RAG, which is one of the most popular modern use cases for Neo4j today."

The Importance of Explainability in AI

As AI systems become more prevalent, explainability and transparency will be crucial for building trust. Eifrem argues that graph-based approaches have an advantage here compared to opaque vector embeddings:

"In graph space, you know that an apple is a fruit and a ball is an object. You know exactly why they're related. An apple and orange are related because they're fruits. That visibility, that literal connection, that transparency leads to explainable decisions."

He continues:

"That transparent way of looking at the world, compared to the opaque nature of the LLM and the vector space, is what gives an AI system trust if you have the more explicit one."

This brings the conversation full circle - from the importance of transparency in building Neo4j as an open source company, to the critical role of transparent, explainable AI systems in the future.

Key Takeaways

• Transparency is fundamental to building trust, both in business and in AI systems
• Open source can be a powerful way to build trust and adoption for new technologies
• Company values and culture need to evolve over time, but in an intentional way that maintains authenticity
• Retrieval is a critical and often overlooked component of modern AI systems
• Graph-based approaches enable more explainable and trustworthy AI

As AI continues to transform industries, companies that prioritize transparency and explainability will be best positioned to build trust with users and customers. Neo4j's journey offers valuable lessons for navigating this new frontier.

Article created from: https://www.youtube.com/watch?v=K93tX3X-SPc