Tackling AI in Texas

We're closely following the issue of AI and emerging technologies and how the State of Texas is considering them, from the standpoint of policies and regulations, to their uses by public and private actors in Texas, to their integration into state agencies, the risks and benefits of AI and emerging technologies, plus how things like ethics, data privacy and cybersecurity factor in. We want to keep you updated on this fast-moving and complex issue, which is why we’ve created this newsletter – Tackling AI in Texas.

Below you’ll find info on an upcoming one-of-a-kind event we’re hosting, plus the news driving the day on AI in Texas.

Data centers vs. Texas power grid

Texas, known for its favorable business climate, is gaining the attention of private equity firms like Blackstone and tech giants such as Microsoft and Alphabet, who want to buy up land for AI expansion thanks to the state’s large land availability and low energy costs.

As a result, the Dallas-Fort Worth area has become the second-biggest US market for leased data center space, causing concern for Texas Senator Nathan Johnson “about data centers and the consumption of power as AI computing becomes part of our everyday life.”

This rapid growth of data centers, cryptocurrency mining, population growth, and other energy-intensive activities has the potential to put a severe strain on the state's power grid, which will need to nearly double its capacity by 2030 to support the anticipated 152 gigawatts of demand on peak days.

The growth even has Lt. Gov. Dan Patrick, often a champion of pro-business policies, hesitant about the effects of data center expansion in the state, remarking on X that data centers and crypto miners “produce very few jobs compared to the incredible demands they place on our grid.”

ERCOT’s demand growth projections led to a joint statement from Gov. Abbott and Lt. Gov. Patrick calling for an immediate review of all policies concerning the grid.

• “All data centers needing more than 75 megawatts of power and hoping to come online in less than two years need to get special approval from ERCOT under rules put in place in 2022.”

Texas is exploring various solutions to meet energy demands, including expanding natural gas capacity and considering future technologies like small modular nuclear reactors. However, these solutions will not be available immediately.

In the face of potential energy restraints, some data center developers are looking at other regions or states. State leaders aren’t looking to get rid of data centers but want to proceed with caution to ensure grid stability.

• “We want data centers, but it can’t be the Wild Wild West of data centers and crypto miners crashing our grid and turning the lights off.” - Lt. Gov. Dan Patrick

When should humans intervene to safeguard against AI bias?

The $184 billion artificial intelligence (AI) market continues to grow, with many organizations, state entities, and companies using AI for decision-making. A 2023 IBM study highlights that 43% of CEOs are utilizing AI for strategic decisions. However, the reliance on AI raises concerns about bias.

One proposed solution to mitigate bias is to make AI systems explain their decision-making processes. The idea is that by reviewing these explanations, human decision-makers can identify and correct biased recommendations. Yet, new research from Texas McCombs challenges the effectiveness of such explanations.

Key points from the research by Maria De-Arteaga and her colleagues include:

• Explanation Types: The study involved AI predicting if individuals were teachers or professors based on 134,436 online biographies. Participants received one of two types of explanations for the AI's decisions: task-relevant keywords (e.g., "research") or gender-related keywords (e.g., "he" or "she").
• Override Likelihood: Participants were 4.5 percentage points more likely to override AI recommendations when explanations highlighted gender, suggesting a heightened sensitivity to perceived gender bias.
• Accuracy Issues: Despite this sensitivity, neither type of explanation improved the accuracy of human overrides compared to participants without explanations. The explanations did not enhance decision quality or fairness.
• Illusion of Fairness: Task-relevant explanations may lead to false assumptions of AI neutrality. Biases can be hidden in task-relevant features, and explanations might not reveal these biases, misleading users into trusting biased recommendations.
• Improvement Suggestions: To enhance the utility of AI explanations the researchers suggest the following:

1. Setting clear objectives for what explanations should be achieved
2. Including fairness-related cues in explanations
3. Providing more insight into the AI’s workings and data limitations
4. Studying how explanations impact human decision-making to align explanations with real-world interactions.

Why it matters: As AI systems become more integrated into the everyday lives of Texans, understanding the tool’s shortcomings will allow users to use them ethically. The research from the McCombs School underscores AI systems as a supplementary tool to human activities, due to a well-documented history of bias, emphasizing the need to not remove humans from the decision-making process.

Event Recap: Texas Tribune’s “Texas and the AI Revolution”

Hosted by the Texas Tribune’s Darla Cameron, Texas and the AI Revolution included discussions from experts Amina Al Sherif, technical engineering lead in machine learning and generative AI at Google; Sherri Greenberg, chair of Good Systems and assistant dean for state and local government engagement at the University of Texas at Austin; and Angela Wilkins, chief data officer at Starling Medical and member of Texas’ Artificial Intelligence Advisory Council on the importance of responsible AI use and collaboration between various stakeholders in business and government.

The panelists suggested that successful AI integration in Texas businesses and government relies on dedicated teams that focus on experimentation and adaptability, moving beyond the initial excitement of the tool often referred to as "shiny new toy syndrome."

The panel emphasized that AI is fundamentally advanced mathematics rather than true Artificial General Intelligence (AGI), and urged for practical, informed approaches to implementation and research, which the panel believed stems from government collaboration with tech companies. Additionally, they discussed the need for the workforce to evolve alongside technological advancements to ensure that there are people in place who can responsibly use these technologies.

When it came time to discuss concerns, environmental concerns were raised, particularly regarding AI's impact on power grids and water usage. Google expert Al Sherif stated that “one ChatGPT question requires the equivalent of 16 ounces of water for the machine to cool down.” With ongoing concerns about the stability of the Texas power grid and Texas water resources, the additional stress brought on by data centers and AI has some calling for regulation now more than ever.

Despite some resistance to adopting AI, the panelists advocated for experimentation to unlock AI's potential to enhance business operations, create new opportunities, and improve quality of life.

Is there anything natural gas can do about AI and data centers’ power demand?

Data center power demand is soaring as AI applications expand, creating a critical need for reliable, 24/7 power, leaving many asking: is natural gas the answer?

Projections indicate that by 2030, U.S. data center energy demand could reach up to 296 terawatt-hours (TWh) annually, depending on the growth scenario. As data centers increasingly rely on AI, their power requirements are surging, making natural gas possibly the most feasible energy source in the near term due to the intermittent nature of renewable energy options like wind and solar.

As of March 2024, the U.S. had 5,381 data centers, nearly doubling from around 2,700 in January 2021. This growth stems largely from the integration of what is considered “AI-powered tasks” — think speech recognition, image recognition, generative AI etc. — into our everyday lives.

Notably, in 2023 alone, nearly 50% of the U.S. data center load was concentrated in only four states: Virginia, Texas, California, and Illinois. For Texas specifically, data centers are drawn in by low energy costs, minimal regulation, and a booming economy.

Texas is the nation's largest electricity consumer and already faces anxiety with grid stability. The increase in data centers will likely being additional challenges. Pablo Vegas, ERCOT CEO, told NPR in an interview that “the amount of energy that it takes to run the AI search is between 10 and 30 times the power requirement to do a traditional Google search.”

Some of the power demand will be satisfied by renewables and renewables-based battery storage, while other traditional fuels will provide some of the generation load. But how much of a natural gas impact can we expect? Probably not as much as one would expect. In the EIA’s most recent Annual Energy Outlook, the agency’s “reference case” showed natural gas demand in decline, primarily due to lower demand from the power sector, with all fossil fuels losing market share to renewables.

The bottom line: While AI is rapidly expanding, it may not lead to a proportional increase in natural gas demand. The data center industry's emphasis on energy efficiency and technological improvements could offset the expected rise in power consumption. As a result, energy suppliers may face a more complex and uncertain market in the future.

Know someone who may want to receive this newsletter?

If you have friends, colleagues or anyone you think might be interested in keeping up with the fast-paced movements of AI and how Texas regulators and policymakers are tackling it, share this link so they can subscribe!