Stanford Tech History Project

Download the full Stanford Tech History Project final report here.

The Stanford PIT Lab wrote a section of the Stanford Tech History Project’s Final Report, which documents how Stanford’s tech ecosystem has changed over the past decade. The Stanford PIT Lab wrote the section about tech recruiting at Stanford, which can be found between pages 17 to 57 of the final report. The section also makes a number of recommendations to improve tech recruiting at Stanford, including ways to create more pathways into PIT jobs and careers.

Artificial Intelligence Explainer

Download the full AI Explainer here.

The Public Interest Technology Lab at Stanford University has developed a series of artificial intelligence (AI) explainers to help bridge the gap between technologists and policymakers. Here, we provide an introduction to AI for policymakers and highlight the key challenges that AI poses to policymakers. We then share our AI Explainer Framework to break down technologies and explain them through visualizations. We applied this framework to explain three particular applications of AI. In the coming weeks, we will release explainers for hiring algorithms, autonomous vehicles, and risk assessment tools.

Hiring Algorithms Explainer

Download the full Hiring Algorithms Explainer here.

Key Takeaways:

• Overview: The hiring process can be broken down into two stages: sourcing and screening. During sourcing, employers often purchase targeted advertising services on social media and job platforms to market jobs to prospective candidates. During screening, employers use tools such as resume-screening technology and chatbots to evaluate candidates.

• Challenges: Advertising algorithms based on user engagement data often reinforce stereotypes in job ads, while screening algorithms with knowledge of a company’s hiring history tend to mimic employers’ discriminatory hiring practices.

• Legislation: Despite recent state and municipal action, the status quo is far from ideal. The Equal Employment Opportunity Commission (EEOC) has not provided much clarity on how policymakers can step in to address ethical concerns in advertising algorithms used in the hiring process. Additionally, auditor independence concerns and the lack of well-defined standards have inhibited the effectiveness of auditing screening tools.

We summarize our breakdown of hiring algorithms in the visualization below.

Risk Assessment Tools Explainer

Download the full Risk Assessment Tools Explainer here.

Key Takeaways:

• Overview: When a defendant is accused of a crime, they are brought before a judge to decide whether they should face pre-trial detention and if so, what the amount of bail should be. Risk assessment algorithms are being increasingly employed by judges to help make this pre-trial detention decision. These algorithms often use historical criminal data, arrest and court records, and background questionnaire responses to assess the likelihood of the defendant reoffending.

• Challenges: Risk assessment tools often lack transparency and perpetuate human bias. The manner in which these tools are deployed can introduce new elements of bias into pre-trial detention decisions. For example, pre-trial officers often have substantial influence over both the data used by risk assessment algorithms and how their results are interpreted.

• Legal approaches: Regulatory measures such as audits can reveal racial bias in the design and deployment of risk assessment tools. The first court test of risk assessment tools, State v. Loomis, largely upheld the legality of the tools but raised questions of fairness and individualization that policymakers will confront.

We summarize our breakdown of risk assessment tools in the visualization below.

Autonomous Vehicles Explainer

Download the full Autonomous Vehicles Explainer here.

Key Takeaways:

• Overview: The process of autonomous driving can be split into three distinct stages: sensing, planning, and control. The sensing stage involves collecting data about the surrounding environment; the planning stage entails interpreting that data to identify the next driving steps for the autonomous vehicle; and the control stage refers to putting these driving steps into action.

• Challenges: In the sensing stage, user data collected by autonomous vehicles can be misused for purposes such as location tracking. In the planning stage, algorithms can more frequently misidentify darker-skinned pedestrians and yield demographic disparities in pedestrian safety. Finally, in the control stage, there is debate regarding the extent to which humans should be allowed to override the decisions of fully autonomous vehicles.

• Legislation: A national regulatory framework for autonomous vehicles is still largely undefined, and federal action has been primarily advisory. For states, the current legislative landscape is a patchwork of exemptions and restrictions across states, varying in both content and scope. Major limitations in this space include uncertainty about the future of autonomous vehicles and debate regarding the balance between safety and innovation.

We summarize our breakdown of autonomous vehicles in the visualization below.