At Intuit, the financial software giant, AI becomes “founding”
Welcome to the special August edition of Eye on AI
For any business, cash is king. This is especially true for small businesses that lack the ability to easily tap into lines of credit, loans, or other types of debt and equity financing available to larger businesses. “I used to run several small businesses and one of the most important things for me was managing money,” says Ashok Srivastava. “Will I have enough money to pay the bills at the end of the month? Can I do payroll?
That’s why Srivastava, who is now the chief data officer of financial software giant Intuit, is especially proud to have helped create an AI-powered cash forecasting tool that many small businesses rely on. now to ensure that they will have enough liquidity. The tool examines outstanding bills, past payment habits, and known upcoming expenses. If the AI software thinks there is likely a shortfall, the system recommends actions the business owner can take to hopefully close the gap or, in some cases, fix the gap. to help get a loan. Over the past year, thanks to improvements to the system by the Srivastava team, the number of people using the cash flow forecasting tool has increased more than fivefold and, in 95% of cases, the business owner takes the necessary actions for the system to recommend.
And if the business turns out to be in need of a loan, the AI helps assess the company’s creditworthiness and ensure they get a decision quickly. Intuit says 60% of customers who get loans through this product, which is called QuickBook Capital, would not qualify for a loan from a bank. “We can provide the loan because we have so much good data and AI,” says Srivastava. He says products like this have saved the lives of many small businesses during the pandemic and will likely continue to do so as many businesses battle soaring inflation.
Srivastava says these are just a few examples of how Intuit is integrating AI-powered features into its products, which include popular tax preparation software TurboTax and QuickBooks, which small businesses often use for accounting, as well as credit monitoring software Credit Karma, personal finance software Mint, and email marketing platform Mailchimp. “This is the direction we are heading,” says Srivastava. “We want to provide advice on a large scale.” He says that means making sure advice is powered by AI “wherever it makes sense” and that AI works to help human experts, not replace them.
Like many companies, Intuit finds that advances in natural language processing are having some of the most significant business impacts. For example, although Intuit makes software, much of its business actually relies on customers consulting with human experts, such as tax consultants or accounting specialists. In the past, these experts had to type notes about the conversation while listening to the customer. This meant that sometimes the experts were not able to spend enough time thinking about the customer’s problem and how to solve it. Now, language AI systems automatically take notes for expert call handlers, automatically summarize the conversation, and classify the caller’s problem. This helps the expert find the best information to convey to the client. But it also helps Intuit track trends and spot potential issues with the company’s software that its developers might need to address.
And Srivastava says society is “increasingly investing” in great language systems, especially ones that can turn speech into text and then parse that text. “Advice comes from the conversation,” he says, so it’s important that AI systems can help bring those conversations to life and analyze their content.
He also says that Intuit is among companies experimenting with reinforcement learning to improve their AI systems, like this cash flow analysis feature in QuickBooks. Reinforcement learning is the type of AI that learns from experience, essentially through trial and error, how to maximize a reward specified by the person creating the AI. In particular, Intuit is exploring the use of a type of reinforcement learning called multi-armed bandit, which attempts to balance the amount of time an AI system explores new options against the time it spends recommending next actions that it has already found effective. It’s AI like this that helps Intuit refine the advice it gives a small business on how best to improve cash flow.
Intuit is an example of a company that places AI at the very heart of its business. “AI is fundamental to our operation,” says Srivastava. It is likely that many other companies will soon find the same thing.
Here are a few other things happening in AI this week.
AI IN THE NEWS
Controversial facial recognition app PimEyes can be used to find images of children, including explicit images. It is the conclusion of a investigation by The Intercept, who used digitally generated images of fake children to search using PimEyes and found that images of many real children came up as possible matches. In many cases, the outlet said, it would have been easy to identify these children. In other cases, the post revealed that the app pulled images of children labeled “potentially explicit.” He said using the app in this way “could lead to increased exploitation of children at a time when the dark web has unleashed an explosion of abuse images”. Giorgi Gobronidze, the owner of PimEyes, told The Intercept that he tasked the app’s engineers with creating better protections for children, but also believes parents need to be more careful and responsible when they posted images of their children online, especially on public websites.
The war in Ukraine has complicated efforts to ban “killer robots”. Deutsche Welle, the German news site, register with the United Nations committee debating what to do about lethal AI-enabled autonomous weapons. International human rights groups, technologists and many countries want these types of weapons banned. But so far the UN has made little progress towards any legally binding restriction on their use, despite eight years of debate. Now, the war in Ukraine has further complicated progress at the UN, the publication reports. One problem is that Russia, which feels diplomatically isolated over its invasion of Ukraine, has claimed that international sanctions are preventing its experts from attending UN committee meetings in Geneva and has used his power to try to block discussions in their absence. But perhaps a bigger problem is that the conflict in Ukraine is, for many military observers, proof of the value of AI-enabled “wandering munitions.” Otherwise known as kamikaze drones, these weapons have been deployed by both sides in the conflict. Current versions have a degree of autonomy, but generally still need a human to select the target they will hit. However, fully autonomous versions are on the horizon.
An AI system learned to use “alternative physics” to explain what it was seeing in the videos. Columbia University researchers wanted to see if an AI system could learn the fundamental variables that underlie physics – phenomena such as mass, velocity, acceleration – by watching simple videos of objects in motion (a swinging pendulum or an inflated “dancer” balloon waving in the wind.) It turned out that the system could actually learn to predict the motion it saw in the videos and could identify and output the number of physical variables that he used to make the prediction. But, in a finding that surprised scientists when they studied what these variables were, they often found that they were apparently different from those used by human scientists to explain and predict the type of movement depicted in videos. In fact, in a few cases, researchers were puzzled as to what the AI was focusing on that allowed it to make accurate predictions. The discovery of this “alternative physics” raises all sorts of questions about whether our current physical model is really the best. “Perhaps some phenomena seem cryptically complex because we’re trying to understand them using the wrong set of variables,” Hod Lipson, director of the Creative Machines Lab in Columbia’s mechanical engineering department, told the publication SciTechDaily.
Our mission to improve business is fueled by readers like you. To enjoy unlimited access to our journalism, subscribe today.