SB 177-AI, DEEPFAKES, CYBERSECURITY, DATA XFERS  3:50:25 PM CO-CHAIR KAWASAKI reconvened the meeting and announced the consideration of SENATE BILL NO. 177 "An Act relating to artificial intelligence; requiring disclosure of deepfakes in campaign communications; relating to cybersecurity; and relating to data privacy." 3:52:11 PM SENATOR SHELLEY HUGHES, District M, Alaska State Legislature, Juneau, Alaska, co-presented an overview as the sponsor of SB 177. She stated that the bill emphasizes transparency, which informs and protects Alaskans. Artificial Intelligence (A.I.) is considered a new technology, but its concept was developed earlier in time. In 1950, Alan Turing wrote a paper titled "Computing Machinery and Intelligence." In 1956, three men put together a program called "Logic Theorist," which mimicked human problem-solving skills, which was presented at a conference titled "Dartmouth Summer Research Project." During that same year, the term 'A.I.' was coined. 3:53:37 PM SENATOR HUGHES said since these milestones, A.I. technology has been perceived as a new frontier, despite having existed for almost 75 years. Its evolution and capabilities have developed at an unprecedented pace. The momentum around this technology is historic and revolutionary, and is becoming a normal component to our lives. Global Positioning System (GPS) technology is assisted by A.I. She has heard emotional expressions ranging from fear and anxiety to anticipation and excitement with the emergence of A.I. technology. As policymakers, it is important to realize that this technology is just another tool. It is also critical to safeguard the public and defend against bad actors. Considering the vast amount of data in state agencies, the state should ensure data outputs are well protected and deepfake is appropriately controlled. 3:56:42 PM SENATOR HUGHES moved to slide 2 of the presentation: [Original punctuation provided.] Defining A.I.  ARTIFICIAL INTELLIGENCE: an automated system that uses data input, human- defined objectives, machine learning, natural language processing, or other computational processing techniques of similar or greater complexity to make a decision or facilitate human decision-making. 3:57:03 PM SENATOR HUGHES moved to slide 3. [Original punctuation provided.]   Why now? Why here? WHY NOW? A.I. is here. It is evolving at lightening speed. We cannot stop it. We cannot ignore it. "A.I. is a tool and in itself is not inherently evil. Our job is to protect against bad actors and harness A.I. for good the very best we can." -Senator Shelley Hughes WHY HERE? Congress is unlikely to unite on parameters and best practices anytime soon. State legislatures are more nimble and ready to mitigate the harm and bridle the benefits of A.I. SENATOR HUGHES said conversations about A.I. are taking place in state capitols across the country, but the U.S. Congress is unlikely to take legislative action on this front. She said she serves on the National Conference of State Legislature's task force on A.I. One advantage of taking legislative action is that the state could more quickly respond and customize. 3:58:20 PM SENATOR HUGHES moved to slide 4: [Original punctuation provided.] Why this focus? 1. State Agency Use of A.I. a) Targeting private sector development and deployment would stifle innovation and be a fool's errand for a state with a small population. b) Setting the parameters for state agency use is necessary 1. to safeguard the public 2. to ensure appropriate deployment that will offer efficiencies and solutions for the workplace 2. Political Deepfakes. a) It's 2024! Elections are around the corner. b) In general, lack of trust chaos. SENATOR HUGHES said because of its small population, Alaska should support innovation rather than targeting vendors. She opined that focusing on state agencies is the right step forward and the public would appreciate doing so. Microsoft Corporation will soon introduce A.I. tools for state government use. Elections are around the corner. She hopes society doesn't have to question reality, but believes it is a possibility that would create chaos. It is important to have the ability to trust one's sense of truth. She expressed concerns about the constitutional republic and the integrity of elections. 4:00:41 PM SENATOR HUGHES moved to slide 5 and highlighted A.I. principles: [Original punctuation provided.]   A Good Starting Point  Agreeing on AI Principles  • Differentiate between tool and actor o Protect against bad actors o Support innovation for beneficial uses • Aim for tech neutrality • Assign human oversight and responsibility • Maintain transparency • Avoid harm/injury • Respect sensitive personal data privacy and security • Embrace data hygiene • Avoid creating/reinforcing unfair bias • Uphold laws and protect individual rights SENATOR HUGHES said after a discussion with the Chief Information Officer of the Alaska State Legislature, she gathered the importance of considering data hygiene. 4:01:32 PM STEPHEN KNOUSE, Staff, Senator Shelley Hughes, Alaska State Legislature, Juneau, Alaska, co-presented an overview of SB 177. He moved to slide 6 and said SB 177 would provide a stop-gap to better identify deepfake usage, which would engender more trust into the public. [Original punctuation provided.] SB 177 -- What it does  1. Adds disclosure statement requirements for political deepfake communications 2. Adds new sections regarding state agency use of artificial intelligence and individuals' data. SB177 3. Adds section to allow individual who suffers harm to bring civil action to superior court 4:02:43 PM SENATOR HUGHES commented that SB 177 is a work in progress and acts as a conversation starter. There is opportunity for legislators to provide amendments. 4:03:54 PM MR. KNOUSE moved to slide 7: [Original punctuation provided.] SB 177 -- What it does  Requires biennial inventory and report of AI systems being used by state agencies published on DOA website. 1. Name and vendor of system 2. General capabilities and uses 3. Whether impact assessment completed prior to implementation Requires biennial impact assessments published on DOA website. 4:04:14 PM MR. KNOUSE moved to slide 8 and briefly explained the intent of the impact assessment: [Original punctuation provided.] SB 177 -- What it does  Impact Assessment  1. System efficacy 2. Human oversight 3. Accountability mechanisms 4. Decision appeals process 5. Benefits, liability, and risks to state 6. Effects on liberty, finances, livelihood, and privacy interests of individuals, including effects from geolocation data use 7. Unlawful discrimination or disparate impact on individual or group 8. Policies and procedures governing process of A.I. system use for consequential decision- making. 4:04:32 PM MR. KNOUSE moved to slide 9 and listed state agency requirements under SB 177: [Original punctuation provided.] SB 177 -- What it does  Requires state agencies to 1. Notify individuals who may be legally or significantly affect 2. Obtain individual's consent before soliciting or acquiring sensitive personal data or sharing data with another state agency* 3. Provide appeals process including manual human review 4. Inform and acquire consent if AI used in hiring interview video 5. When outsource, multi-factor authentication must secure system and stored data 4:05:00 PM SENATOR HUGHES added feedback on the final point on slide 9. She said for data to be shared with agencies such as the Department of Public Safety (DPS) or the courts, consent would not be required. She noted that the Alaska State Legislature's Chief Information Officer informed her of the 42 applications on the MyAlaska website that are shared across numerous departments, so the corresponding section under SB 177 will need to be refined. 4:05:40 PM MR. KNOUSE moved to slide 10 and listed rules for state agencies under SB 177: [Original punctuation provided.] SB 177 -- What it does  Prohibits state agencies from using.  1.Biometric identification e.g., facial recognition 2.Emotion recognition 3.Cognitive behavioral manipulation of individuals or groups 4.Social scoring 5.AI systems that use data hosted in hostile nations. 4:06:01 PM SENATOR HUGHES moved to slide 11 and summarized state agency provisions: [Original punctuation provided.] Process Checks and Balances  • Transparent  o What are we doing (algorithmic assessment made public) • Explainable  o Why are we doing it • Conservative  o Within scope and purpose only • Accountable  o Human oversight o Developer o Deployer o User o Regular audits/assessments, and retraining as-needed o Secure data storage SENATOR HUGHES said when looking at A.I., oversight is important at all stages in its development. Narrowing the scope to generative A.I. would support checks and balances and reduce the fiscal requirements. She noted that opposed to generative A.I., rule-based A.I. is more like an Excel spreadsheet where data is entered and a result is calculated. It would be beneficial to narrow A.I. definitions. The fiscal note for SB 177 is large because it includes rule-based A.I. Restricting the language of SB 177 to focus on generative A.I. would greatly reduce the fiscal note. 4:08:08 PM CHAIR KAWASAKI wondered if either private or publicly-held companies would have the ability to store biometric identification. 4:08:46 PM SENATOR HUGHES replied that multi-factor authentication would be required to protect data within state agency contracts under SB 177. She offered to obtain further information from DPS. 4:09:37 PM MR. KNOUSE moved to slide 12 and explained deepfakes in elections: [Original punctuation provided.] Deepfakes in Elections  Generative A.I. can convincingly imitate elected leaders and other public figures. AI tools can synthesize audio in any person's voice and generate realistic still and moving visuals of almost anyone doing anything. The proliferation of "synthetic media" poses challenges to the functioning of our constitutional republic when such communications can deprive the public of the accurate information it needs to make informed decisions in elections. 4:10:16 PM MR. KNOUSE moved to slide 13 and listed examples: [Original punctuation provided.] Deepfakes in Elections  • USA: Voice altered to sound like Biden urged voters in New Hampshire not to cast ballots. • TURKEY: News outlet published deepfake video showing a party endorsing opposition; presidential candidate withdrew from race due to a deepfake "sex tape" video • SLOVAKIA: Leader Michal ?imecka was depicted as saying he would raise the price of beer and had plans to rig the election. 4:10:52 PM MR. KNOUSE moved to slide 14 and read the following: [Original punctuation provided.] Deepfakes in Elections  "The fact-checkers trying to hold the line against disinformation on social media in Slovakia say their experience shows AI is already advanced enough to disrupt elections, while they lack the tools to fight back." 4:11:10 PM MR. KNOUSE moved to slide 15 and presented A.I. generated photos of Joe Biden and Donald Trump. 4:11:36 PM MR. KNOUSE moved to slide 16 and presented an artificially-made news video segment about Santa Claus. 4:12:30 PM SENATOR HUGHES noted that a free A.I. tool was used for the video example, but paid software would be more realistic. 4:12:42 PM MR. KNOUSE moved to slide 17 and showcased a deepfake-generated video clip of senators whose faces were swapped with Star Wars characters. 4:13:44 PM MR. KNOUSE moved to slide 18 and showed a deepfake-generated video of Senator Hughes as Queen Elizabeth II. He reiterated that more robust software would be far more convincing. 4:14:13 PM SENATOR HUGHES said the videos she created were made to be intentionally less convincing to prevent widespread circulation, but the ability to create more realistic content is possible. She stated that one presidential candidate in the recent presidential election in Turkey used deepfake technology to falsely distort his opponent's character and uplift his own appearance. A candidate running for public office has the ability to create deepfake content falsifying one's own character. 4:15:10 PM SENATOR HUGHES moved to slide 19 and highlighted potential solutions: [Original punctuation provided.] Solutions  • Power of disclosure o Establish norms, standards, and laws now o People can discount untruths • Enforcement and penalties are necessary • Injunctive relief is important SENATOR HUGHES stated that SB 177 does not include injunctive relief, but while a court would likely order a deepfake video be removed, it would be possible for content to already been copied and shared on the Internet. 4:17:04 PM At ease. 4:17:45 PM CHAIR KAWASAKI reconvened the meeting. 4:18:27 PM CHAIR KAWASAKI announced invited testimony. 4:19:20 PM SPENSE PURNELL, Director, Technology Policy, Reason Foundation, Tallahassee, Florida, invited testimony for SB 177. He expressed disappointment knowing there is no existing federal framework on the issue of A.I. regulation, but acknowledged that other states are stepping in to fill the policy-void. He opined that SB 177 would do several positive things and addresses state use of AI, which could help identify threats in addition to opportunities. He said the third section in the bill regarding civil liability is an essential piece and expressed anticipation to see what state agencies could accomplish with this bill. SB 177 could also address potential threats that have not yet been realized and it underlines the idea that A.I. is a quickly-evolving technology. He conveyed that there are a range of actions states could take to navigate A.I. technology, such as banning it entirely or crafting legislation to manage its use. He stated his belief that SB 177 may require refinement, but appreciates that it is less invasive. He encouraged the committee to research and learn about A.I. technology to help define parts of the bill. 4:24:23 PM CHAIR KAWASAKI asked if Adobe Photoshop and its filtering tool falls under the definition of a deepfake given its computational processing technique that allows programs to alter one's appearance. 4:25:05 PM MR. PURNELL replied that the example of Adobe Photoshop demonstrates the downside to establishing a broad definition. Google Search and tools such as spellcheck or could be arguably defined as A.I. technology. The computational processing technique is likely too broad and would need to be refined in statute. He opined that the state may benefit from taking inventory of A.I. uses, but current language under SB 177 could be refined. 4:26:39 PM SENATOR HUGHES clarified that the computational processing technique is included in part of the deepfake definition under SB 177. She expressed the importance of remaining technology neutral. She stated that when she previously worked on unmanned aircraft policy, aviation rules were generalized, so bill language on deepfake could include provisions on manually- produced or photo-shopped material in addition to generative A.I. She suggested updating the bill language so definitions are technology-neutral. However, state agency usage may require a closer lens on generative A.I usage. 4:29:07 PM DANIEL CASTRO, Director, Data Innovation, Information Technology & Innovation Foundation, Washington, District of Columbia, testified in support of SB 177 and provided recommendations. He said there have been rapid advances in generative A.I. technology, so the public is rightfully concerned that bad actors could exploit it to create deepfake content. Deepfake technology has the potential to harm and influence elections, harm a candidate's reputation, mislead voters, and foster mistrust. He made several suggestions to reduce the risk of deepfake exploitation: 1. Update state election laws to make it unlawful for campaigns or political organizations to knowingly distribute materially-deceptive media that uses a person's likeness to manipulate voters without providing clear and conspicuous disclosure. 2. A.I. generated content should allow beneficial uses, such as a simple campaign video edit, when done with the candidate's consent. 3. Avoid requirements for disclosure that inadvertently mandate disclaimers on all media, which could desensitize voters to these types of nuisances. 4. Warning the public on deceptive media rather than explicitly pointing out A.I. uses, which is more informative and less stigmatizing. 5. Developing a robust enforcement mechanism that requires swift consequences and penalties for distributing deepfake content. 6. The legislative focus should center on political organizations that are intentionally creating or distributing deepfakes; it is important to avoid creating a patchwork of conflicting state laws for various platforms. 4:33:27 PM MR. CASTRO said he appreciated the committee continuing the discussion on this issue as Election Day nears. The growth rate and ease of of A.I. technology usage is escalating, so he urged the committee to address public concerns without constraining many of its beneficial uses. 4:34:01 PM CHAIR KAWASAKI asked about the current and potential approach to enforcement. 4:34:35 PM MR. CASTRO replied that it is important to differentiate genuine from illegitimate campaigns, especially in recognizing the influence of deepfake used by Russian hackers to sway elections. While federal and state laws are unable to prevent these types of activities, legislative action could prevent legitimate campaigns from using robocalls, creating deepfake content, or falsifying advertisements. Developing a swift response and establishing civil penalties would effectively mitigate the influence of A.I. on elections. 4:36:37 PM NATE PERSILY, Professor, Stanford Law School, Stanford, California, invited testimony for SB 177. He emphasized that A.I. has had its "Taylor Swift moment," which entailed the publishing and circulation of a deepfake photo of the singer. This incident acted as an opening silo and its impact will be apparent through political expression. The use of deepfake technology will blossom and multiply quickly. He suggested the following: 1. Identify dimensions or subjects to regulate, and focus on creators and speakers, outlined in SB 177. 2. Ensuring regulations are not placed on standardized Internet communication tools do not establish measures that would detrimentally impact the political process. 3. Identify the location of regulation. A law could affect deepfake creation anywhere in the U.S. on the Internet. 4. Narrow a definition for organic communication versus advertising. 5. Define libelous deepfakes or efforts intended to damage political parties or candidates. 6. Develop criminal or civil liability enforcement measures. MR. PERSILLY encouraged members to review the EU AI Act, which comprises a series of definitions developed by the European Union that the state could potentially modify for particular context. The approach to regulating deepfake content varies among different states, while SB 177 specifically measures damage done to the reputation of a political candidate. 4:41:19 PM MR. PERSILY said that for political campaigns, it is important to determine when and to whom regulations would apply. First amendment concerns currently exist around vagueness, content- based, and viewpoint-based discrimination. To address this, SB 177 focuses on the disclaimer mandate for communications. For example, Citizens United struck down campaign finance laws but upheld disclaimer requirements. Defining deepfake context is important because disclosure measures would not raise all the same constitutional concerns as general libel. He stated his belief that this focus area will occupy the legislature. 4:43:15 PM SENATOR WIELECHOWSKI stated his belief that SB 177 only requires a disclaimer if deepfake content injures the reputation of a candidate or party, so the interpretation could be subjective. He wondered if a better policy would comprise a direct disclaimer. 4:43:50 PM MR. PERSILY replied that part of the challenge is that the narrower disclaimer is more likely to be upheld. Every current digital platform utilizes A.I., so there is a risk of disclaimers applying to all imagery on the Internet and people would become numb to disclaimers to any content with an A.I. tag. He stated his belief that the rationale is to limit disclaimers to the types of deepfakes people are currently worried about. One way to avoid the viewpoint-based discrimination claim is to enlarge the focus on political communications. He suggested using Stanford Law School's libel language for reckless regard if the legislature decided to focus on reputational injury. 4:46:34 PM ROBERT WEISSMAN, President, Public Citizen, Washington, District of Columbia, invited testimony for SB 177 on behalf of Public Citizen. He thanked Senator Hughes and the committee for their leadership. He conveyed that while it is important to get the details of deepfake regulations right, quick action is necessary. Failing to act would result in damaging consequences for Alaska and around the world. The political deepfake moment is upon society. The examples seen in the U.S. and around the world are harbingers of what will come in the absence of regulation. He guarantees that political operatives of all backgrounds will misuse A.I. technology without rules in place or legal repercussions, and the impact would stretch beyond foreign influence on elections. This could result in a late-game shift in election influence, widespread voter doubt, and candidates denying the truth. 4:48:22 PM MR. WEISSMAN suggested that the state must take action. Public Citizen asserts that the most important thing is to establish a law and social norms that define political deepfakes without a disclosure as unacceptable. He opined that it is crucial to establish the standards contained in SB 177 and maintained that a disclosure mandate is the correct approach to take because it takes into consideration the first amendment and Supreme Court jurisprudence. He recommended strengthening SB 177 by doing the following: 1. Focus on enforcement and strengthen legal repercussions. 2. Establish a right to conjunctive relief; an impacted candidate has every incentive for quick action. 3. Specify that the use of deepfakes for satire would not be covered under the bill to mitigate first amendment challenges. MR. WEISSMAN said states across the country are taking action on this issue. Although there is bipartisan support for federal legislation to regulate deepfake content, Congress is unlikely to take timely action. 4:51:31 PM CHAIR KAWASAKI asked if any states have gotten ahead of the game to deploy A.I. legislation to prevent the potential concerns mentioned for 2024. 4:51:59 PM MR. WEISSMAN said there are five states that have passed laws prohibiting deepfakes without disclosure, including Michigan, and more than two dozen states have introduced similar bills. 4:52:36 PM CHAIR KAWASAKI referenced the New Hampshire deepfake robocall scenario that did not clearly state the call was from "Joe Biden," but sounded like him. He said legislation in Alaska would not have penalized or prevented this type of incident from happening. He asked how the legislature could prevent this situation from happening or whether it could establish a penalty. 4:53:30 PM MR. WEISSMAN replied that Public Citizen would strongly favor that approach. He believes there are current laws in Alaska that prohibit the dissuasion of voting, which are established in New Hampshire. Even in the absence of a deepfake bill in that state, the robocall incident was illegal. He suggested adding a specific measurement concerning the use of deepfakes that deter or mislead voters. 4:55:13 PM SENATOR HUGHES said while SB 177 is a work in progress, it is important to start the conversation. She noted a section concerning state agency use and data that will need to be updated. The Alaska State Legislature's Chief Information Officer informed her about a policy that refers to a dynamic list on state departments, so including this component would be wise. 4:56:01 PM CHAIR KAWASAKI stated that there is a lot to consider on the issue of A.I. 4:56:36 PM CHAIR KAWASAKI held SB 177 in committee.