We in Telegram
Add news
March 2010 April 2010 May 2010 June 2010 July 2010
August 2010
September 2010 October 2010
November 2010
December 2010
January 2011
February 2011 March 2011 April 2011 May 2011 June 2011 July 2011 August 2011 September 2011 October 2011 November 2011 December 2011 January 2012 February 2012 March 2012 April 2012 May 2012 June 2012 July 2012 August 2012 September 2012 October 2012 November 2012 December 2012 January 2013 February 2013 March 2013 April 2013 May 2013 June 2013 July 2013 August 2013 September 2013 October 2013 November 2013 December 2013 January 2014 February 2014 March 2014 April 2014 May 2014 June 2014 July 2014 August 2014 September 2014 October 2014 November 2014 December 2014 January 2015 February 2015 March 2015 April 2015 May 2015 June 2015 July 2015 August 2015 September 2015 October 2015 November 2015 December 2015 January 2016 February 2016 March 2016 April 2016 May 2016 June 2016 July 2016 August 2016 September 2016 October 2016 November 2016 December 2016 January 2017 February 2017 March 2017 April 2017 May 2017 June 2017 July 2017 August 2017 September 2017 October 2017 November 2017 December 2017 January 2018 February 2018 March 2018 April 2018 May 2018 June 2018 July 2018 August 2018 September 2018 October 2018 November 2018 December 2018 January 2019 February 2019 March 2019 April 2019 May 2019 June 2019 July 2019 August 2019 September 2019 October 2019 November 2019 December 2019 January 2020 February 2020 March 2020 April 2020 May 2020 June 2020 July 2020 August 2020 September 2020 October 2020 November 2020 December 2020 January 2021 February 2021 March 2021 April 2021 May 2021 June 2021 July 2021 August 2021 September 2021 October 2021 November 2021 December 2021 January 2022 February 2022 March 2022 April 2022 May 2022 June 2022 July 2022 August 2022 September 2022 October 2022 November 2022 December 2022 January 2023 February 2023 March 2023 April 2023 May 2023 June 2023 July 2023 August 2023 September 2023 October 2023 November 2023 December 2023 January 2024 February 2024 March 2024 April 2024 May 2024 June 2024
1 2 3 4 5 6 7 8 9 10 11 12 13 14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
News Every Day |

Words to Fear: I’m From the State Government, and I’m Here to Help with AI Risk

Jack Solowey and Jennifer Huddleston

State legislatures want to help with AI risk.

California lawmakers’ heavy‐​handed attempt to regulate cutting‐​edge AI development (SB 1047) received appropriate attention and backlash from the pro‐​innovation policy community. That backlash shone an important spotlight on the challenges state legislatures pose to AI innovation nationally.

Indeed, 40 states have considered some form of AI legislation this year alone, threatening to create an unworkable multi‐​state patchwork. The initiatives range from laws targeting specific AI applications (such as AI’s use in music under Tennessee’s ELVIS Act) to regulatory regimes for AI broadly.

In addition to legislation like SB 1047 designed to tackle frontier AI risk, another category of relatively broad AI legislation worth paying close attention to seeks to regulate AI’s use in so‐​called “consequential decisions,” such as employment, health care, and financial determinations.

While the risks these consequential‐​decision acts seek to tackle are seemingly more mundane than the putative threats of AI‐​enabled mass destruction targeted by frontier model legislation, the risks the consequential‐​decision acts pose to AI development are by no means trivial. They threaten AI development by undermining naturally emerging business models, putting AI developers in jeopardy, and targeting technologies not just harms.

Consequential‐​decision acts are gaining traction in the states. On May 17, Democratic Governor Jared Polis signed into law Colorado’s consequential‐​decision bill, albeit under a bit of protest. With the Colorado bill advancing into law before even the EU’s AI Act, the typically innovation‐​embracing state has come out ahead in a race it shouldn’t want to win.

On May 21, the California State Assembly passed its variation on the theme in a bill covering “Automated decision tools” (AB 2930), which is now before the California State Senate.

Though their details vary, both the Colorado and California consequential‐​decision acts seek to combat the risks of AI decision tools perpetrating algorithmic discrimination (which is roughly defined to mean unlawful disparate treatment or impacts disfavoring people based on their membership in a protected class). Specifically, the acts seek to combat discrimination when AI tools are used for decisions that have material, legal, or similar effects on access to things like education, employment, housing, utilities, health care, legal services, and financial services.

The Colorado and California regulatory approach addresses potential discrimination through a suite of obligations placed to varying extents on AI decision tool developers and deployers (i.e., the organizations using the tools in interactions with consumers). The acts generally impose duties—differing in their particulars—to avoid algorithmic discrimination, perform risk assessments, notify individuals regarding the use of AI decision tools, provide consumers with rights to opt out of and/​or appeal automated decisions and implement AI governance programs designed to mitigate the risks of algorithmic discrimination.

The Colorado and California acts’ automated decision opt‐​out and appeal rights provide a window into the two regimes’ similarities, subtle differences, and ultimate problems. Whereas the California bill creates a new automated decision opt‐​out right on top of an existing one mentioned in the state’s consumer privacy law, the Colorado law refers to a similar right in the state’s own privacy law while also adding new rights to appeal certain automated decisions.

Notably, California’s data privacy regulator has also begun preliminary rulemaking activity for California’s existing automated decision opt‐​out right. This points to a broader conversation that must be had regarding the interactions between data privacy laws and AI. Existing privacy regulations may not be well‐​adapted to the AI era. For instance, such laws’ data minimization requirements and limits on the use of personal information could undermine attempts to combat bias through more diverse data sets.

artificial intelligence

As for the opt‐​out/​appeal rights in the automated‐​decision acts themselves, generally, both bills require some form of alternative decision process or human review when it’s requested by a consumer and is “technically feasible,” but Colorado would require the consumer to wait for an adverse decision, while California is less clear on timing.

There’s something superficially enticing to many about circumventing automated decisions, but creating a blanket right to do so is not without costs. Indeed, automation often will be precisely what provides the cost savings that allow a business to offer products or services at an attractive price. Mandated opt‐​out rights likely would result in certain products and services becoming more expensive or unavailable.

Colorado’s more limited appeal right is a better approach but ultimately would impose similar costs, just to a potentially lesser degree. Furthermore, the caveat in the acts that alternative processes and human review be “technically feasible” is unlikely to help businesses with the technical ability to provide alternatives but without the resources to do so cost‐​effectively.

Absent opt‐​out mandates, businesses still would be able to provide such rights in response to consumer demand, while the broader ecosystem could simultaneously provide a greater range of features and prices.

The opt‐​out mandates’ constraint on naturally emerging business models is one of the core issues with the Colorado and California proposals. The others are the legal jeopardy and compliance burdens imposed on AI developers, as well as regulatory approaches that target technologies instead of harms.

The Colorado and California consequential‐​decision acts both impose onerous compliance risks and obligations on AI developers. Specifically, the acts inappropriately require developers, not just deployers, to anticipate and mitigate the risks of algorithmic discrimination. (Absurdly, the California bill even obligates AI developers to give legal advice to deployers, requiring developers to provide a “description of the deployer’s responsibilities under” the act.)

AI artificial

One major problem with this general approach is that it’s difficult, if not impossible, for a developer to completely understand an AI system’s propensity for discrimination in a vacuum or to predict every possible way their tool may be used. Any discriminatory effect of an AI system likely would be a product of both the underlying model and the deployer’s use, including the real‐​world data the deployer feeds into the model at the inference stage, as well as the deployer’s ability to implement compensating controls addressing any disparate outputs.

One way the acts address this problem is by cabining some developer obligations to only those risks that are “reasonably foreseeable.” Nonetheless, the California bill undermines this limitation by imposing a general duty on developers to “not make available” an AI decision tool “that results in algorithmic discrimination.” While the Colorado law does a better job of limiting developer duties to only reasonably foreseeable risks, it nonetheless has unreasonable expectations regarding what developers will be able to predict and take responsibility for. The Colorado law mistakenly assumes developers will have greater knowledge of an AI tool’s “intended uses” than is likely to be the case and requires developers to notify law enforcement after discovering their tool’s use by a deployer is likely to have caused algorithmic discrimination.

Requiring developers to orient their compliance measures around predicted use cases risks limiting the types of productive ends to which their models may be applied, as novel use cases could increase compliance risk. Disincentivizing developers from allowing all but the most obvious intended uses would be a huge loss for the AI ecosystem, as some of the most creative applications of technologies typically are devised downstream from the tool’s creator. That’s why, for example, third‐​party apps exist for smartphones.

Perhaps the original sin of the consequential‐​decision acts is that they target AI used for, well, consequential decisions. Such decisions tend to be those related to sectors that already are heavily regulated, such as health care and finance. For example, the core risk addressed by these acts—discrimination based on protected class membership—already is illegal in credit decisions under federal law. Targeting the technology, as opposed to the harm, in the financial services context, for instance, is redundant at best and counterproductive at worst, as it adds yet another layer of compliance burden that could stymie AI tools’ potential to expand credit access to the historically underserved. In addition, this general approach often misassigns the blame for bad or negligent actors’ improper use of technology to the technology itself.

This shortsighted regulatory playbook—constraining business models, burdening developers with responsibility for downstream risks, and targeting technologies instead of harms—is being employed all too often at the state level. After all, SB 1047 is a notorious vehicle for all three, making open‐​source AI development a compliance risk by requiring developers to lock down their models against certain downstream modifications, as well as targeting technical sophistication, not merely specific threats.

The risk from this playbook is that the US will be made worse off as state‐​level frameworks become de facto national standards without the benefit of national input. This is not just the case for legislation out of large states like California, as laws with long‐​arm ambitions and cloud‐​based targets can, in practice, extend compliance burdens beyond state borders. Where that’s the case, conflicting obligations and subtle variations can raise the question of whether full‐​scale compliance is even possible.

Instead of seeking to be the first to regulate, states should consider working from an alternative playbook that prioritizes innovation, avoids counterproductive interventions, and targets harms, not technologies. In the meantime, we should fear the words, “I’m from the state government, and I’m here to help with AI risk,” even when it’s another state’s government saying them.

Москва

Филиал № 4 ОСФР по Москве и Московской области информирует: Свыше 5,2 миллиона жителей Московского региона получают набор социальных услуг в натуральном виде

America’s Space Infrastructure: So Vulnerable It Destabilizes Geopolitics

Kai Havertz’s fiancee Sophia stuns in see-through outfit on hen weekend as fans say Arsenal star is ‘so lucky’

Creepy King Charles Painting Is Vandalized by Radical Animal Rights Activists, Gets a Cartoon Face and Speech Balloon!

Panthers take Game 1 of Stanley Cup Final as Sergei Bobrovsky shuts out Oilers

Ria.city






Read also

Luka Doncic’s terrible defense is an anchor weighing down Mavs in NBA Finals

South African parties hold crunch talks to decide who will govern

British Olympic dressage entries, an exciting arrival, and more things the horse world is talking about

News, articles, comments, with a minute-by-minute update, now on Today24.pro

News Every Day

Panthers take Game 1 of Stanley Cup Final as Sergei Bobrovsky shuts out Oilers

Today24.pro — latest news 24/7. You can add your news instantly now — here


News Every Day

Panthers take Game 1 of Stanley Cup Final as Sergei Bobrovsky shuts out Oilers



Sports today


Новости тенниса
Ролан Гаррос

Испанец Алькарас выиграл Открытый чемпионат Франции по теннису



Спорт в России и мире
Москва

Массовыми патриотическими акциями отметили День России на заводах АО "Желдорреммаш"



All sports news today





Sports in Russia today

Москва

Сотрудники Росгвардии приняли участие в спортивном празднике МГО «Динамо» в Москве


Новости России

Game News

MMORPG Tarisland выпустили в Китае раньше времени


Russian.city


Москва

Золотые часы от Министерства обороны за вклад в развитие Кадетского движения России!


Губернаторы России
Владимир Путин

Путин приветствовал в России участников и гостей Спортивных игр стран БРИКС


В первый день Игр БРИКС Россия завоевала 13 золотых медалей

Соревнования по мотокроссу прошли в Химках

«СВЯТОЙ ЛЕНИН» и ДЕНЬ РОССИИ: успех России в С В О = мир во всём мире?!

Российский рынок акций – разум победил эмоции


Музыкальная сборная России сыграет в Воронцовском дворце шедевры Рахманинова и Баха

«Приносит деньги!»: Бутман заявил о нераздельности культуры и экономики

Рэпер ST встретился с участниками «Российской студенческой весны» в Ставрополе

Концерт ко Дню России провели в ДК «Родина» в Химках


Испанский теннисист Надаль не сыграет на Уимблдоне

Мирра Андреева может стать новой Марией Шараповой

Хачанов вылетел с турнира ATP в Хертогенбосхе после первого же матча

Соболенко в купальнике зажигательно станцевала перед женатым бойфрендом. Видео



Золотые часы от Министерства обороны за вклад в развитие Кадетского движения России!

Филиал № 4 ОСФР по Москве и Московской области информирует: Свыше 5,2 миллиона жителей Московского региона получают набор социальных услуг в натуральном виде

Массовыми патриотическими акциями отметили День России на заводах АО "Желдорреммаш"

Владимир Путин, Дмитрий Медведев, Анатолий Голод: ПРИЗНАНИЕ НАТО НАДО ПРИМЕНЯТЬ!


Парадоксы двойного стандарта

Стали известны итоги жеребьёвки Пути РПЛ Кубка России по футболу

Путин заявил, что Россия преодолеет любые испытания

Почему у канализационной системы должен быть выход на крышу?


Часть банков в Киргизии перестала принимать деньги из России

Сотрудники «Мособлпожспаса» вывели заблудившуюся женщину из леса под Шаховской

Южная Корея нацелилась на Чёрный континент

Масштабный праздник двора организовали в Химках



Путин в России и мире






Персональные новости Russian.city
Филипп Киркоров

«Не кричи!»: Елена Север и Филипп Киркоров представили совместный клип и трек



News Every Day

Creepy King Charles Painting Is Vandalized by Radical Animal Rights Activists, Gets a Cartoon Face and Speech Balloon!




Friends of Today24

Музыкальные новости

Персональные новости