Monthly Archives: May 2023

In 2019, Apples credit card business came under fire for offering a woman one twentieth the credit limit offered to her husband. When she complained, Apple representatives reportedly told her, I dont know why, but I swear were not discriminating. Its just the algorithm.

Today, more and more decisions are made by opaque, unexplainable algorithms like this often with similarly problematic results. From credit approvals to customized product or promotion recommendations to resume readers to fault detection for infrastructure maintenance, organizations across a wide range of industries are investing in automated tools whose decisions are often acted upon with little to no insight into how they are made.

This approach creates real risk. Research has shown that a lack of explainability is both one of executives most common concerns related to AI and has a substantial impact on users trust in and willingness to use AI products not to mention their safety.

And yet, despite the downsides, many organizations continue to invest in these systems, because decision-makers assume that unexplainable algorithms are intrinsically superior to simpler, explainable ones. This perception is known as the accuracy-explainability tradeoff: Tech leaders have historically assumed that the better a human can understand an algorithm, the less accurate it will be.

Specifically, data scientists draw a distinction between so-called black-box and white-box AI models: White-box models typically include just a few simple rules, presented for example as a decision tree or a simple linear model with limited parameters. Because of the small number of rules or parameters, the processes behind these algorithms can typically be understood by humans.

In contrast, black-box models use hundreds or even thousands of decision trees (known as random forests), or billions of parameters (as deep learning models do), to inform their outputs. Cognitive load theory has shown that humans can only comprehend models with up to about seven rules or nodes, making it functionally impossible for observers to explain the decisions made by black-box systems. But does their complexity necessarily make black-box models more accurate?

To explore this question, we conducted a rigorous, large-scale analysis of how black and white-box models performed on a broad array of nearly 100 representative datasets (known as benchmark classification datasets), spanning domains such as pricing, medical diagnosis, bankruptcy prediction, and purchasing behavior. We found that for almost 70% of the datasets, the black box and white box models produced similarly accurate results. In other words, more often than not, there was no tradeoff between accuracy and explainability: A more-explainable model could be used without sacrificing accuracy.

This is consistent with other emerging research exploring the potential of explainable AI models, as well as our own experience working on case studies and projects with companies across diverse industries, geographies, and use cases. For example, it has been repeatedly demonstrated that COMPAS, the complicated black box tool thats widely used in the U.S. justice system for predicting likelihood of future arrests, is no more accurate than a simple predictive model that only looks at age and criminal history. Similarly, a research team created a model to predict likelihood of defaulting on a loan that was simple enough that average banking customers could easily understand it, and the researchers found that their model was less than 1% less accurate than an equivalent black box model (a difference that was within the margin of error).

Of course, there are some cases in which black-box models are still beneficial. But in light of the downsides, our research suggests several steps companies should take before adopting a black-box approach:

As a rule of thumb, white-box models should be used as benchmarks to assess whether black-box models are necessary. Before choosing a type of model, organizations should test both and if the difference in performance is insignificant, the white-box option should be selected.

One of the main factors that will determine whether a black-box model is necessary is the data involved. First, the decision depends on the quality of the data. When data is noisy (i.e., when it includes a lot of erroneous or meaningless information), relatively simple white-box methods tend to be effective. For example, we spoke with analysts at Morgan Stanley who found that for their highly noisy financial datasets, simple trading rules such as buy stock if company is undervalued, underperformed recently, and is not too large worked well.

Second, the type of data also affects the decision. For applications that involve multimedia data such as images, audio, and video, black-box models may offer superior performance. For instance, we worked with a company that was developing AI models to help airport staff predict security risk based on images of air cargo. They found that black-box models had a higher chance of detecting high-risk cargo items that could pose a security threat than equivalent white-box models did. These black-box tools enabled inspection teams to save thousands of hours by focusing more on high-risk cargo, substantially boosting the organizations performance on security metrics. In similarly complex applications such as face-detection for cameras, vision systems in autonomous vehicles, facial recognition, image-based medical diagnostic devices, illegal/toxic content detection, and most recently, generative AI tools like ChatGPT and DALL-E, a black box approach may be advantageous or even the only feasible option.

Transparency is always important to build and maintain trust but its especially critical for particularly sensitive use cases. In situations where a fair decision-making process is of utmost importance to your users, or in which some form of procedural justice is a requirement, it may make sense to prioritize explainability even if your data might otherwise lend itself to a black box approach, or if youve found that less-explainable models are slightly more accurate.

For instance, in domains such as hiring, allocation of organs for transplant, and legal decisions, opting for a simple, rule-based, white-box AI system will reduce risk to both the organization and its users. Many leaders have discovered these risks the hard way: In 2015, Amazon found that its automated candidate screening system was biased against female software developers, while a Dutch AI welfare fraud detection tool was shut down in 2018 after critics decried it as a large and non-transparent black hole.

An organizations choice between white or black-box AI also depends on its own level of AI readiness. For organizations that are less digitally developed, in which employees tend to have less trust in or understanding of AI, it may be best to start with simpler models before progressing to more complex solutions. That typically means implementing a white-box model that everyone can easily understand, and only exploring black-box options once teams have become more accustomed to using these tools.

For example, we worked with a global beverage company that launched a simple white-box AI system to help employees optimize their daily workflows. The system offered limited recommendations, such as which products should be promoted and how much of different products should be restocked. Then, as the organization matured in its use of and trust in AI, managers began to test out whether more complex, black-box alternatives might offer advantages in any of these applications.

In certain domains, explainability might be a legal requirement, not a nice-to-have. For instance, in the U.S., the Equal Credit Opportunity Act requires financial institutions to be able to explain the reasons why credit has been denied to a loan applicant. Similarly, Europes General Data Protection Regulation (GDPR) suggests that employers should be able to explain how candidates data has been used to inform hiring decisions. When organizations are required by law to be able to explain the decisions made by their AI models, white-box models are the only option.

Finally, there are of course contexts in which black-box models are both undeniably more accurate (as was the case in 30% of the datasets we tested in our study) and acceptable with respect to regulatory, organizational, or user-specific concerns. For example, applications such as computer vision for medical diagnoses, fraud detection, and cargo management all benefit greatly from black-box models, and the legal or logistical hurdles they pose tend to be more manageable. In cases like these, if an organization does decide to implement an opaque AI model, it should take steps to address the trust and safety risks associated with a lack of explainability.

In some cases, it is possible to develop an explainable white-box proxy to clarify, in approximate terms, how a black-box model has reached a decision. Even if this explanation isnt fully accurate or complete, it can go a long way to build trust, reduce biases, and increase adoption. In addition, a greater (if imperfect) understanding of the model can help developers further refine it, adding more value to these businesses and their end users.

In other cases, organizations may truly have very limited insight into why a model makes the decisions it does. If an approximate explanation isnt possible, leaders can still prioritize transparency in how they talk about the model both internally and externally, openly acknowledging the risks and working to address them.

***

Ultimately, there is no one-size-fits-all solution to AI implementation. All new technology comes with risks, and the choice of how to balance those risks with the potential rewards will depend on the specific business context and data. But our research demonstrates that in many cases, simple, interpretable AI models perform just as well as black box alternatives without sacrificing the trust of users or allowing hidden biases to drive decisions.

The authors would like to acknowledge Gaurav Jha and Sofie Goethals for their contribution.

Read the rest here:

AI Can Be Both Accurate and Transparent - HBR.org Daily

In the spring of 2007, I was one of four journalists anointed by Steve Jobs to review the iPhone. This was probably themost anticipated product in the history of tech. What would it be like? Was it a turning point for devices? Looking back atmy review today, I am relieved to say its not an embarrassment: I recognized the devices generational significance. But for all the praise I bestowed upon the iPhone, I failed to anticipate its mind-blowing secondary effects, such as the volcanic melding of hardware, operating system, and apps, or its hypnotic effect on our attention. (I did urge Apple to encourage outside developers to create new uses for the device.) Nor did I suggest we should expect the rise of services like Uber or TikTok or make any prediction that family dinners would turn into communal display-centric trances. Of course, my primary job was to help people decide whether to spend $500, which was super expensive for a phone back then, to buy the damn thing. But reading the review now, one might wonder why I spent time griping about AT&Ts network or the web browsers inability to handle Flash content. Thats like quibbling over what sandals to wear just as a three-story tsunami is about to break.

I am reminded of my failure of foresight when reading about the experiences people are having with recent AI apps, likelarge language model chatbots and AIimage generators. Quite rightfully, people are obsessing about the impact of a sudden cavalcade ofshockingly capable AI systems, though scientists often note that these seemingly rapid breakthroughs have been decades in the making. But as when I first pawed the iPhone in 2007, we risk failing to anticipate the potential trajectories of our AI-infused future by focusing too much on the current versions of products like Microsofts Bing chat, OpenAIs ChatGPT, Anthropics Claude, and Googles Bard.

This fallacy can be clearly observed in what has become a new and popular media genre, best described as prompt-and-pronounce. The modus operandi is to attempt some task formerly limited to humans and then, often disregarding the caveats provided by the inventors, take it to an extreme. The great sports journalist Red Smith once said that writing a column is easyyou just open a vein and bleed. But would-be pundits now promote a bloodless version: You just open a browser and prompt. (Note: this newsletter was produced the old-fashioned way, by opening a vein.)

Typically, prompt-and-pronounce columns involve sitting down with one of these way-early systems and seeing how well it replaces something previously limited to the realm of the human. In a typical example, aNew York Times reporter used ChatGPT toanswer all her work communications for an entire week.The Wall Street Journals product reviewer decided toclone her voice (hey,we did that first!) and appearance using AI to see if her algorithmic doppelgngers could trick people into mistaking the fake for the real thing. There are dozens of similar examples.

Generally, those who stage such stunts come to two conclusions: These models are amazing, but they fall miserably short of what humans do best. The emails fail to pick up workplace nuances. The clones have one foot dragging in the uncanny valley. Most damningly, these text generators make things up when asked for factual information, a phenomenon known as hallucinations' that is the current bane of AI. And its a plain fact that the output of todays models often have a soulless quality.

In one sense, its scarywill our future world be run by flawed mind children, as roboticist Hans Moravec calls our digital successors? But in another sense, the shortcomings are comforting. Sure, AIs can now perform a lot of low-level tasks and are unparalleled at suggesting plausible-looking Disneyland trips and gluten-free dinner party menus, butthe thinking goesthe bots will always need us to make corrections and jazz up the prose.

See original here:

You're Probably Underestimating AI Chatbots | WIRED - WIRED

WASHINGTON (AP) Computer engineers and tech-inclined political scientists have warned for years that cheap, powerful artificial intelligence tools would soon allow anyone to create fake images, video and audio that was realistic enough to fool voters and perhaps sway an election.

The synthetic images that emerged were often crude, unconvincing and costly to produce, especially when other kinds of misinformation were so inexpensive and easy to spread on social media. The threat posed by AI and so-called deepfakes always seemed a year or two away.

No more.

Sophisticated generative AI tools can now create cloned human voices and hyper-realistic images, videos and audio in seconds, at minimal cost. When strapped to powerful social media algorithms, this fake and digitally created content can spread far and fast and target highly specific audiences, potentially taking campaign dirty tricks to a new low.

The implications for the 2024 campaigns and elections are as large as they are troubling: Generative AI can not only rapidly produce targeted campaign emails, texts or videos, it also could be used to mislead voters, impersonate candidates and undermine elections on a scale and at a speed not yet seen.

Were not prepared for this, warned A.J. Nash, vice president of intelligence at the cybersecurity firm ZeroFox. To me, the big leap forward is the audio and video capabilities that have emerged. When you can do that on a large scale, and distribute it on social platforms, well, its going to have a major impact.

AI experts can quickly rattle off a number of alarming scenarios in which generative AI is used to create synthetic media for the purposes of confusing voters, slandering a candidate or even inciting violence.

Here are a few: Automated robocall messages, in a candidates voice, instructing voters to cast ballots on the wrong date; audio recordings of a candidate supposedly confessing to a crime or expressing racist views; video footage showing someone giving a speech or interview they never gave. Fake images designed to look like local news reports, falsely claiming a candidate dropped out of the race.

What if Elon Musk personally calls you and tells you to vote for a certain candidate? said Oren Etzioni, the founding CEO of the Allen Institute for AI, who stepped down last year to start the nonprofit AI2. A lot of people would listen. But its not him.

Former President Donald Trump, who is running in 2024, has shared AI-generated content with his followers on social media. A manipulated video of CNN host Anderson Cooper that Trump shared on his Truth Social platform on Friday, which distorted Coopers reaction to the CNN town hall this past week with Trump, was created using an AI voice-cloning tool.

A dystopian campaign ad released last month by the Republican National Committee offers another glimpse of this digitally manipulated future. The online ad, which came after President Joe Biden announced his reelection campaign, and starts with a strange, slightly warped image of Biden and the text What if the weakest president weve ever had was re-elected?

A series of AI-generated images follows: Taiwan under attack; boarded up storefronts in the United States as the economy crumbles; soldiers and armored military vehicles patrolling local streets as tattooed criminals and waves of immigrants create panic.

An AI-generated look into the countrys possible future if Joe Biden is re-elected in 2024, reads the ads description from the RNC.

The RNC acknowledged its use of AI, but others, including nefarious political campaigns and foreign adversaries, will not, said Petko Stoyanov, global chief technology officer at Forcepoint, a cybersecurity company based in Austin, Texas. Stoyanov predicted that groups looking to meddle with U.S. democracy will employ AI and synthetic media as a way to erode trust.

What happens if an international entity a cybercriminal or a nation state impersonates someone. What is the impact? Do we have any recourse? Stoyanov said. Were going to see a lot more misinformation from international sources.

AI-generated political disinformation already has gone viral online ahead of the 2024 election, from a doctored video of Biden appearing to give a speech attacking transgender people to AI-generated images of children supposedly learning satanism in libraries.

AI images appearing to show Trumps mug shot also fooled some social media users even though the former president didnt take one when he was booked and arraigned in a Manhattan criminal court for falsifying business records. Other AI-generated images showed Trump resisting arrest, though their creator was quick to acknowledge their origin.

Legislation that would require candidates to label campaign advertisements created with AI has been introduced in the House by Rep. Yvette Clarke, D-N.Y., who has also sponsored legislation that would require anyone creating synthetic images to add a watermark indicating the fact.

Some states have offered their own proposals for addressing concerns about deepfakes.

Clarke said her greatest fear is that generative AI could be used before the 2024 election to create a video or audio that incites violence and turns Americans against each other.

Its important that we keep up with the technology, Clarke told The Associated Press. Weve got to set up some guardrails. People can be deceived, and it only takes a split second. People are busy with their lives and they dont have the time to check every piece of information. AI being weaponized, in a political season, it could be extremely disruptive.

Earlier this month, a trade association for political consultants in Washington condemned the use of deepfakes in political advertising, calling them a deception with no place in legitimate, ethical campaigns.

Other forms of artificial intelligence have for years been a feature of political campaigning, using data and algorithms to automate tasks such as targeting voters on social media or tracking down donors. Campaign strategists and tech entrepreneurs hope the most recent innovations will offer some positives in 2024, too.

Mike Nellis, CEO of the progressive digital agency Authentic, said he uses ChatGPT every single day and encourages his staff to use it, too, as long as any content drafted with the tool is reviewed by human eyes afterward.

Nellis newest project, in partnership with Higher Ground Labs, is an AI tool called Quiller. It will write, send and evaluate the effectiveness of fundraising emails - all typically tedious tasks on campaigns.

The idea is every Democratic strategist, every Democratic candidate will have a copilot in their pocket, he said.

___

Swenson reported from New York.

___

The Associated Pressreceives support from several private foundations to enhance its explanatory coverage of elections and democracy. See more about APs democracy initiative here. The AP is solely responsible for all content.

___

Follow the APs coverage of misinformation at https://apnews.com/hub/misinformation and coverage of artificial intelligence at https://apnews.com/hub/artificial-intelligence

Originally posted here:

AI presents political peril for 2024 with threat to mislead voters - The Associated Press

Readers respond to Naomi Kleins article that argued it is delusional to believe AI machines will benefit humanity

Fri 12 May 2023 11.55 EDT

Naomi Kleins article about the dangers of generative AI makes many valid points about the economic and social consequences of the new technology (AI machines arent hallucinating. But their makers are, 8 May). But her choice of language about how to describe the mistakes that the new AI makes seems to suggest she is committed mainly to providing an ideological interpretation of the new technology.

Saying that mistakes are the results of glitches in the code rather than the tech hallucinating suggests the simulation is a simple one, involving a kind of power of the false rather than a more complex one that allows the possibility of some form of fabulation. This is important because it means that the technology cant be seen simply as a control technology, like nuclear fusion or self-driving cars, but instead indicates a switch to an adaptive form of technology, ie, ones that are based on adapting what is already out there rather than trying to reinvent what exists, as in some form of innovation.

Obviously, climate change will require more of the adaptive kinds of technology, like reusable space rockets and wind farms, because control technologies are very resource heavy and tend to cause a lot of collateral damage.Terry PriceLondon

Naomi Klein is right to voice scepticism about the claims made for generative AI. As its development coincides with endgame capitalism, a minimum requirement for its effective governance must be that those responsible for its programming are truly representative, not only of humanity as a whole but the living planet.

Rather than a group of white, male, wealthy individuals developing AI in their image, we need to ensure that indigenous wisdom, the aspirations of future generations drawn from all continents and those able to identify the impact of potential decisions and actions on our ecosystems all need to participate in the design of these AI developments. Without such input, all such AI will do is exacerbate our demise: with these contributions, it may yet avert it. Surely this is an issue that is too important to be left to Silicon Valley to self-determine.Dave HunterBristol

The real danger of AI systems arises from the fact that these systems have no actual intelligence and so cannot distinguish whether the results they produce are correct or not. ChatGPT produces intelligent results in the midst of a whole lot of other results which, to our human intelligence, are simply ridiculous. This doesnt matter too much because we simply laugh at and discard the ridiculous results.

But when these AI systems are controlling cars and planes, where the ridiculous results are a danger to life and cant just be discarded, the consequences could be catastrophic. The artificial neural networks producing AI are bandied about as emulators of the brain. But in spite of decades of dedicated research, neural networks have just 10 to 1,000 neurons, whereas the human brain has 86bn of them.

No wonder that an AI system has no way of knowing whether it has produced an intelligent (by human standards) result.Charles RoweWantage, Oxfordshire

It is understandable that there is concern over the effect that AI will have on our future, but I am equally concerned about the damage that humans will do if were left in charge (Why the godfather of AI fears for humanity, 5 May).

Would an AI system really have dealt with the Covid pandemic worse than Boris Johnson? Would it have allowed our planet to get so close to the precipice of climate catastrophe? Geoffrey Hinton believes that once AI is more intelligent than us, it will inevitably take charge, and perhaps he is right to be concerned. On the other hand, it might be just what we need.Ben ChesterStroud, Gloucestershire

Have an opinion on anything youve read in the Guardian today? Please email us your letter and it will be considered for publication in our letters section.

{{topLeft}}

{{bottomLeft}}

{{topRight}}

{{bottomRight}}

{{.}}

Continued here:

We need AI to help us face the challenges of the future - The Guardian

Updated May 15, 2023, 04:03pm EDT

Loop Capital downgraded its stock rating for Google parent Alphabet from a buy to a hold in a buzzy Monday morning note to clients, throwing some cold water on last weeks boundless optimism for the Google parents future in the budding artificial intelligence sphere.

The boom in AI chatbots could cause "behavioral changes" that make users less likely to rely on traditional search engines, placing a "ceiling" on Alphabet's valuation, Loop Capital analysts Rob Sanderson and Alan Gould wrote in a Monday note.

The strategists set a $118 target for Alphabet shares, a tick above the stocks $117 price Monday and well below the $130 average analyst target for the stock, according to FactSet.

Its the first downgrade for Alphabet since at least March 3, per Factset.

Shares of Alphabet tumbled nearly 1% in Monday trading, moving against the tech-heavy Nasdaqs slight gains, though the stock is still up more than 10% over the past week as investors cheered on the companys Wednesday presentation outlining the incorporation of AI into various phases of its business.

Loops skepticism on Alphabet comes not from concerns about the oft-discussed gains made by Microsoft in the AI space but rather about how generative AI chatbots like Microsoft-backed ChatGPT call into question Googles long-standing status as the primary gateway to the web, calling the shift among users a competitive force against its dominance in connecting users to information.

Even if AI caps Alphabets upside, it remains one of the largest companies on earth, with its $1.4 trillion market capitalization trailing only those of Apple, Microsoft and Saudi Aramco.

Google has unmatched AI competencies and will be a major beneficiary of AI adoption over the long-term, Sanderson and Gould clarified. That echoes the bullish sentiments expressed by many analysts following Googles I/O developer conference; firms such as Bank of America, Goldman Sachs, UBS and JPMorgan reiterated their buy ratings for Alphabet in notes late last week.

Sanderson and Gould published another note early Monday upgrading Metas stock from a hold to a buy, setting a $320 price target for the Facebook and Instagram parent, indicating 37% upside in what would be the stocks highest price since January 2022. Shifting macroeconomic tides and faith in Instagram Reels and Metas AI push in advertising all serve as tailwinds for the stock, according to the strategists. Meta shares rallied more than 2% in Monday trading.

Google Insiders Are $9 Billion Richer After AI-Fueled Stock Rally (Forbes)

Google Adding AI To Search Engine For Some UsersClosing In On Microsoft's AI Push (Forbes)

Googles Peacetime CEO Sundar Pichai Faces Criticism As The AI War Heats Up (Forbes)

I'm a New Jersey-based Senior Reporter on our news desk. I graduated in 2021 from Duke University, where I majored in Economics and served as sports editor for The Chronicle, Duke's student newspaper. Send tips at dsaul@forbes.com.

Go here to read the rest:

End Of Googles Dominance? Stock Gets Rare Analyst Downgrade Over AI Fears - Forbes

The most accurate simulation of objects made from tens of millions of atoms has been run on one the worlds top supercomputers with the help of artificial intelligence.

Existing simulations that describe in detail how atoms behave, interact and evolve are limited to small molecules, because of the computational power needed. There are techniques to simulate much larger numbers of atoms through time, but these rely on approximations and arent accurate enough to extract many detailed features of the molecule in question.

Now, Boris Kozinsky at Harvard University and his colleagues have developed a tool, called Allegro, that can accurately simulate systems with tens of millions of atoms using artificial intelligence.

Kozinsky and his team used the worlds 8th most powerful supercomputer, Perlmutter, to simulate the 44 million atoms involved in the protein shell of HIV. They also simulated other common biological molecules such as cellulose, a protein missing in people with haemophilia and a widespread tobacco plant virus.

Anything thats essentially made out of atoms, you can simulate with these methods at extremely high accuracy, and now also at large scale, says Kozinsky. This is one demonstration, but by no means constrained to this domain. The system could also be used for many problems in materials science, such as investigating batteries, catalysis and semiconductors, he says.

To be able to simulate such large numbers of particles, the researchers used a kind of AI called a neural network to calculate interactions between atoms that were symmetrical from every angle, a principle called equivariance.

When you develop networks that very fundamentally include these symmetries you get these big improvements in accuracy and other properties that we care about, such as the stability of simulations, or how fast the machine learning model learns as you teach it with more data, says team member Albert Musaelian, also at Harvard.

This is a tour de force in programming and demonstrating that these machine-learned potentials are now scalable, says Gbor Csnyi at the University of Cambridge.

However, simulating biological molecules like these is more of a demonstration that the tool works for large systems than a practical boost for researchers, as biochemists already have accurate enough tools that can be run much faster, he says. Where it could be useful is for materials with lots of atoms that experience shocks and extreme forces over very short timescales, such as in planetary cores, says Csnyi.

Topics:

Excerpt from:

Watch 44 million atoms simulated using AI and a supercomputer - New Scientist