Leemon Baird Discusses Hedera’s Technical Strategy and the Future of AI

Leemon Baird introduced his concept of hashgraph consensus in 2016, offering a novel alternative to conventional blockchain frameworks. Drawing from extensive experience in both academic and industrial sectors, he established Hedera to bring this technology to market.

His academic background showcases significant early contributions to neural networks and reinforcement learning in the 1990s, a notable era marked by what is now termed the "AI winter." The Hedera initiative has since grown amid a competitive field of distributed ledger technologies, each asserting its own technical advantages and targeting various market segments.

At Consensus 2025, Baird adeptly navigates between technical discourse and business strategy, illustrating the complexities of both developing groundbreaking technology and fostering a sustainable ecosystem.

Discerned from an interview, adjustments made for clarity.

CoinDesk: Your hashgraph algorithm surfaced in 2016 amidst a wave of alternative consensus mechanisms. What specific shortcomings did you aim to overcome?

Baird: My passion lies in computer science and mathematics—the thrill of innovation and problem-solving. As I ventured into entrepreneurship a quarter-century ago, I maintained a similar approach. The essence of my work centers on identifying the core issue we seek to resolve. In the blockchain realm, I recognized that while Bitcoin was an exciting development, it had limitations in speed and security, lacking in ABFT [Asynchronous Byzantine Fault Tolerance]. The energy consumption was considerable, and its flexibility was insufficient. I questioned whether there was a way to enhance consensus efficiency at its foundational level without excessive energy expenditure while remaining fast and secure. Could we attain ABFT’s utmost security while minimizing our carbon footprint?

My exploration began in 2012, initially doubting its feasibility. I would revisit the problem, only to convince myself it was insurmountable again and again. However, by 2015, I discovered that incorporating two hashes seamlessly aligned everything—allowing for internet-speed transactions coupled with top-tier ABFT security, all while relying on proof of stake, thus conserving energy.

From a business standpoint, I questioned what governance model would be best suited. Many blockchains claim decentralized governance, but power tends to centralize over time, often resulting in a small group of developers or hidden individuals gaining control.

In contrast, we initiated Hedera with a decentralized governance model from the outset. We engaged some of the most reputable organizations worldwide—respected universities and companies—to provide a balanced governance structure, creating necessary checks and balances.

Our approach directly addressed the question: What do you genuinely desire from governance? What fosters a higher degree of trustlessness or fundamentally reduces the trust level needed to fully rely on the system? We applied the same analytical rigor to business challenges as we did to mathematical ones.

CoinDesk: You’ve indicated that RWA tokenization, carbon credits, and stablecoins are crucial applications. How has Hedera demonstrated transaction volumes or user engagement in these areas?

Baird: I’d pinpoint four primary sectors:

Firstly, the excitement surrounding AI is palpable. However, the risks associated with AI also compel us to establish provenance, governance, and version control for AI systems. Ensuring trust in AI-generated outcomes is imperative. Hedera facilitates this in various ways, including data permissioning and potentially managing royalties for provided training data. The collaboration between EQTY Lab, NVIDIA, and Intel on Hedera is particularly promising.

Secondly, real asset tokenization is revolutionizing asset management. Many projects on Hedera now tokenize diverse assets like real estate, precious metals, carbon credits, and emissions. From the inception of blockchain technology, I’ve insisted that the ultimate goal is to bring every valuable entity onto these ledgers, rather than focusing solely on trivial collectibles or games.

Thirdly, stablecoins are vital for real-world implementation. We have established a Stable Coin Studio to streamline stablecoin development on Hedera, with numerous financial organizations making substantial strides in this space.

Lastly, Hedera’s Consensus Service offers unparalleled immutable data records, allowing for access-controlled message distribution while ensuring permanent record-keeping. Companies like Hyundai and Kia use this technology for tracking emissions throughout their supply chain.

CoinDesk: UCL recently researched energy consumption across various networks. Given the significance of methodology in these assessments, do you consider Hedera’s approach fundamentally different from other PoS networks, or is the efficiency mainly a result of current network setup?

Baird: We’ve always prioritized energy efficiency in our design. From our algorithms to node governance, and our choice of proof of stake over proof of work, each aspect was engineered to minimize emissions from the start.

This approach initiated a positive feedback loop. Early adopters focused on tokenizing carbon credits gravitated towards a green blockchain. As organizations aimed to tokenize emissions and credits, they naturally chose a platform already utilized by others engaged in similar efforts. Consequently, Hedera has emerged as a leading blockchain in the green technology sector.

According to recent findings from University College London, Hedera consistently reports the lowest carbon emissions per transaction across blockchains. We even invest in carbon credits to maintain a carbon-negative standing. Our eco-friendly framework was integrated from day one, reinforcing our leadership in this niche.

CoinDesk: Numerous projects are attempting to merge AI and blockchain technologies. What genuine integration opportunities do you perceive beyond mere promotional claims?

Baird: The intersection of AI and blockchain is more profound than many realize. On Hedera, we’re witnessing tangible advancements in several domains:

First, provenance and governance are critical. As AI-generated content proliferates, it’s essential to confirm the authenticity of that content through digital signatures that establish origin, whether created by a human or an AI.

Secondly, data permissioning plays a crucial role. When myriad contributors supply tiny data chunks for AI training, it’s vital that each individual retains control over their data, including permission for its use.

Looking ahead, I am particularly eager to harness Hedera for identity management, integrating it with AI systems. As we approach a threshold where AI-produced media becomes indistinguishable from real-life content, relying on digital signatures becomes essential. To validate these signatures, we will require robust identity systems.

CoinDesk: Having explored neural networks in the 1990s before the current AI renaissance, how do you view today’s large language models? Is there a fundamental shift in technology, or does it simply reflect advancements in scale?

Baird: Many AI developments have unfolded in ways I anticipated. For instance, I expected the triumph of AlphaGo over the world champion of Go, AlphaZero’s mastery of chess, and AIs excelling at poker—all achieved using techniques I had envisioned. The speed of computing we now possess was a key enabler.

The progress in self-driving cars aligns with my expectations as well, utilizing methodologies I had forecasted.

However, I’ve been utterly amazed by the emergence of ChatGPT and large language models (LLMs). The breakthrough gained from the 2017 transformer architecture—as discussed in "Attention is All You Need"—was something unforeseen. In the 90s, we struggled extensively with language processing, yet now it’s a reality.

The prowess of present-day LLMs continues to astonish me, and their future trajectory is uncertain. Will we see them achieve superintelligence, or will they reach a plateau? That remains unclear, and I would argue that no one truly knows.

Even humanoid robots have exceeded expectations. While their physical advancements align with my projections, their conversational capabilities, especially powered by LLMs, far surpass my initial imaginings. In the near future, they’re likely to start with basic factory tasks before evolving to skilled professions such as welding and plumbing.

These technological strides are poised to make the Industrial Revolution seem trivial in comparison. Many have yet to comprehend the scale of these transformations or their rapid advancement.