Connect with us

AI

Oh dear… AI models used to flag hate speech online are, er, racist against black people

Avatar

Published

on

The internet is filled with trolls spewing hate speech, but machine learning algorithms can’t help us clean up the mess.

A paper from computer scientists from the University of Washington, Carnegie Mellon University, and the Allen Institute for Artificial Intelligence, found that machines were more likely to flag tweets from black people than white people as offensive. It all boils down to the subtle differences in language. African-American English (AAE), often spoken in urban communities, is peppered with racial slang and profanities.

But even if they contain what appear to be offensive words, the message itself often isn’t abusive. For example, the tweet “I saw him yesterday” is scored as 6 per cent toxic, but it suddenly skyrockets to 95 per cent for the comment “I saw his ass yesterday”. The word ass may be crude, but when used in that context it’s not aggressive at all.

racial_bias_hate_speech_detection

An example of how African-American English (AAE) is mistakenly classified as offensive compared to standard American English. Image credit: Sap et al.

“I wasn’t aware of the exact level of bias in Perspective API–the tool used to detect online hate speech–when searching for toxic language, but I expected to see some level of bias from previous work that examined how easily algorithms like AI chatter bots learn negative cultural stereotypes and associations,” said Saadia Gabriel, co-author of the paper and a PhD student at the University of Washington.

“Still, it’s always surprising and a little alarming to see how well these algorithms pick up on toxic patterns pertaining to race and gender when presented with large corpora of unfiltered data from the web.”

The researchers fed a total of 124,779 tweets collected from two datasets that were classified as toxic according to Perspective API. Originally developed by Google and Jigsaw, an incubator company currently operating under Alphabet, the machine learning software is used by Twitter to flag any abusive comments.

The tool mistakenly classified 46 per cent of non-offensive tweets crafted in the style of African American English (AAE) as inflammatory, compared to just nine per cent of tweets written in standard American English.

“I think we have to be really careful about what technologies we implement in general, whether it’s a platform where people can post whatever they want, or whether is an algorithm that detects certain types of (potentially harmful) content. Platforms are under increasing pressure to delete harmful content, but currently these deletions are backfiring against minorities,” Maarten Sap, first author of the paper and a PhD student at the University of Washington, told The Register.

When humans were employed via the Amazon Mechanical Turk service to look at 1,351 tweets from the same dataset and asked to judge if the comment was either offensive to them or could be seen as offensive to anyone.

Just over half – about 55 per cent – were classified as “could be offensive to anyone”. That figure dropped to 44 per cent, however, when they were asked to consider the user’s race and their use of AAE.

shock_computer

Q. If machine learning is so smart, how come AI models are such racist, sexist homophobes? A. Humans really suck

READ MORE

“Our work serves as a reminder that hate speech and toxic language is highly subjective and contextual,” said Sap.

“We have to think about dialect, slang and in-group versus out-group, and we have to consider that slurs spoken by the out-group might actually be reclaimed language when spoken by the in-group.”

The study provides yet another reminder that AI models don’t understand the world enough to have common sense. Tools like Perspective API often fail when faced with subtle nuances in human language or even incorrect spellings.

Similar models employed by other social media platforms like Facebook to detect things like violence or pornography often don’t work for the same reason. And this is why these companies can’t rely on machines alone, and have to hire teams of human contractors to moderate questionable content.

Sap believes that removing the humans from content moderation isn’t the way to go.

“We managed to reduce some of the bias by making workers more aware of the existence of African American English, and reminding them that certain seemingly obscene words could be harmless depending on who speaks them. Knowing how flawed humans are at this task, especially given the working conditions that some companies put their content moderators in, I certainly don’t think humans are flawless in this capacity. However, I don’t think removing them from the equation is necessarily the way to go either. I think a good collaborative human+AI setting is likely the best option, but only time will tell.” ®

Sponsored: Detecting cyber attacks as a small to medium business

Source: https://go.theregister.co.uk/feed/www.theregister.co.uk/2019/10/11/ai_black_people/

Continue Reading

AI

Father and son duo take on global logistics with Optimal Dynamics’ sequential decision AI platform

Avatar

Published

on

Like “innovation,” machine learning and artificial intelligence are commonplace terms that provide very little context for what they actually signify. AI/ML spans dozens of different fields of research, covering all kinds of different problems and alternative and often incompatible ways to solve them.

One robust area of research here that has antecedents going back to the mid-20th century is what is known as stochastic optimization — decision-making under uncertainty where an entity wants to optimize for a particular objective. A classic problem is how to optimize an airline’s schedule to maximize profit. Airlines need to commit to schedules months in advance without knowing what the weather will be like or what the specific demand for a route will be (or, whether a pandemic will wipe out travel demand entirely). It’s a vibrant field, and these days, basically runs most of modern life.

Warren B. Powell has been exploring this problem for decades as a researcher at Princeton, where he has operated the Castle Lab. He has researched how to bring disparate areas of stochastic optimization together under one framework that he has dubbed “sequential decision analytics” to optimize problems where each decision in a series places constraints on future decisions. Such problems are common in areas like logistics, scheduling and other key areas of business.

The Castle Lab has long had industry partners, and it has raised tens of millions of dollars in grants from industry over its history. But after decades of research, Powell teamed up with his son, Daniel Powell, to spin out his collective body of research and productize it into a startup called Optimal Dynamics. Father Powell has now retired full-time from Princeton to become Chief Analytics Officer, while son Powell became CEO.

The company raised $18.4 million in new funding last week from Bessemer led by Mike Droesch, who recently was promoted to partner earlier this year with the firm’s newest $3.3 billion fundraise. The company now has 25 employees and is centered in New York City.

So what does Optimal Dynamics actually do? CEO Powell said that it’s been a long road since the company’s founding in mid-2017 when it first raised a $450,000 pre-seed round. We were “drunkenly walking in finding product-market fit,” Powell said. This is “not an easy technology to get right.”

What the company ultimately zoomed in on was the trucking industry, which has precisely the kind of sequential decision-making that father Powell had been working on his entire career. “Within truckload, you have a whole series of uncertain variables,” CEO Powell described. “We are the first company that can learn and plan for an uncertain future.”

There’s been a lot of investment in logistics and trucking from VCs in recent years as more and more investors see the potential to completely disrupt the massive and fragmented market. Yet, rather than building a whole new trucking marketplace or approaching it as a vertically-integrated solution, Optimal Dynamics decided to go with the much simpler enterprise SaaS route to offer better optimization to existing companies.

One early customer, which owned 120 power units, saved $4 million using the company’s software, according to Powell. That was a result of better utilization of equipment and more efficient operations. They “sold off about 20 vehicles that they didn’t need anymore due to the underlying efficiency,” he said. In addition, the company was able to replace a team of ten who used to manage trucking logistics down to one, and “they are just managing exceptions” to the normal course of business. As an example of an exception, Powell said that “a guy drove half way and then decided he wanted to quit,” leaving a load stranded. “Trying to train a computer on weird edge events [like that] is hard,” he said.

Better efficiency for equipment usage and then saving money on employee costs by automating their work are the two main ways Optimal Dynamics saves money for customers. Powell says most of the savings come in the former rather than the latter, since utilization is often where the most impact can be felt.

On the technical front, the key improvement the company has devised is how to rapidly solve the ultra-complex optimization problems that logistics companies face. The company does that through value function approximation, which is a field of study where instead of actually computing the full range of stochastic optimization solutions, the program approximates the outcomes of decisions to reduce compute time. We “take in this extraordinary amount of detail while handling it in a computationally efficient way,” Powell said. That’s where we have really “wedged ourselves as a company.”

Early signs of success with customers led to a $4 million seed round led by Homan Yuen of Fusion Fund, which invests in technically-sophisticated startups (i.e. the kind of startups that take decades of optimization research at Princeton to get going). Powell said that raising the round was tough, transpiring during the first weeks of the pandemic last year. One corporate fund pulled out at the last minute, and it was “chaos ensuing with everyone,” he said. This Series A process meanwhile was the opposite. “This round was totally different — closed it in 17 days from round kickoff to closure,” he said.

With new capital in the bank, the company is looking to expand from 25 employees to 75 this year, who will be trickling back to the company’s office in the Flatiron neighborhood of Manhattan in the coming months. Optimal Dynamics targets customers with 75 trucks or more, either fleets for rent or private fleets owned by companies like Walmart who handle their own logistics.

Coinsmart. Beste Bitcoin-Börse in Europa
Source: https://techcrunch.com/2021/05/18/optimal-dynamics-series-a/

Continue Reading

AI

IBM will buy Salesforce partner Waeg to boost hybrid cloud, AI strategy

Avatar

Published

on

Elevate your enterprise data technology and strategy at Transform 2021.


(Reuters) — IBM said on Tuesday it would buy Waeg, a consulting partner for Salesforce, in a deal that will extend its range of services and support its hybrid cloud and artificial intelligence strategy.

The deal to acquire Waeg, which is based in Brussels and serves clients across Europe, complements IBM’s acquisition in January of 7Summits, a U.S. consultancy that specialises in Salesforce’s customer management software.

“Waeg’s strength in Salesforce consulting services will be key to creating intelligent workflows that allow our clients to keep pace with changing customer and employee needs and expectations,” Mark Foster, senior vice president of IBM Services and Global Business Services, said.

Waeg employs a team of 130 ‘Waegers’ in locations that include Belgium, Denmark, France, Ireland, Poland, Portugal and the Netherlands.

The terms were not disclosed for the deal, which is subject to customary closing conditions and is expected to be completed this quarter.

VentureBeat

VentureBeat’s mission is to be a digital town square for technical decision-makers to gain knowledge about transformative technology and transact. Our site delivers essential information on data technologies and strategies to guide you as you lead your organizations. We invite you to become a member of our community, to access:

  • up-to-date information on the subjects of interest to you
  • our newsletters
  • gated thought-leader content and discounted access to our prized events, such as Transform 2021: Learn More
  • networking features, and more

Become a member

Coinsmart. Beste Bitcoin-Börse in Europa
Source: https://venturebeat.com/2021/05/18/ibm-will-buy-salesforce-partner-waeg-to-boost-hybrid-cloud-ai-strategy/

Continue Reading

AI

For companies that use ML, labeled data is the key differentiator

Avatar

Published

on

AI is driving the paradigm shift that is the software industry’s transition to data-centric programming from writing logical statements. Data is now oxygen. The more training data a company gathers, the brighter will its AI-powered products burn.

Why is Tesla so far ahead with advanced driver assistance systems (ADAS)? Because no one else has collected as much information — it has data on more than ten billion driven miles, helping it pull ahead of competition like Waymo, which has only about 20 million miles. But any company that is considering using machine learning (ML) cannot overlook one technical choice: supervised or unsupervised learning.

There is a fundamental difference between the two. For unsupervised learning, the process is fairly straightforward: The acquired data is directly fed to the models, and if all goes well, it will identify patterns.

Elon Musk compares unsupervised learning to the human brain, which gets raw data from the six senses and makes sense of it. He recently shared that making unsupervised learning work for ADAS is a major challenge that hasn’t been solved yet.

Supervised learning is currently the most practical approach for most ML challenges. O’Reilly’s 2021 report on AI Adoption in the Enterprise found that 82% of surveyed companies use supervised learning, while only 58% use unsupervised learning. Gartner predicts that through 2022, supervised learning will remain favored by enterprises, arguing that “most of the current economic value gained from ML is based on supervised learning use cases”.

Coinsmart. Beste Bitcoin-Börse in Europa
Source: https://techcrunch.com/2021/05/18/for-companies-that-use-ml-labeled-data-is-the-key-differentiator/

Continue Reading

AI

My Experience Building a WhatsApp Chat Bot for a Nigerian Company

Avatar

Published

on

Jerry Udensi
Summary.

Short introduction: I’m Jerry Udensi, CTO of a Nigerian-Malaysian tech company: Lyshnia Limited. Prior to working full time with Lyshnia (a company I founded in 2013 with my elder brother), I worked in the AI industry in Malaysia and Singapore. I have built Natural Language AI systems for large corporations such as Allianz SE, and Insurance Technology for companies like Malaysia’s Insuradar Sdn.

The reason for my short introduction is to show you my background in building AI powered systems. Natural Language Processing is a field I’ve actively been in for over 3 years now so you’d think building a Transactional Chat Bot that sells only 10 products shouldn’t be an issue for me right? Well you’d be right if the customers were people who read.

In the paragraphs to follow, I will highlight what I’ve learnt building and maintaining Jane B(Just another Non-Existent Bot) which attends to approx. 1000 customers every day.

There’s this old saying that goes “if you want to hide something from a Black Man, put it in a book”. Unfortunately, this is the case with over 70% of the Customers who used the bot.

When you first message the bot, it greets you, let’s you know that you’re chatting with a Bot, then gives you 4 options to choose from.

The first 3 messages you receive after chatting the first time.

5 out of 10 people ignore the initial message and go ahead to write what they want, 2 out of 10 people would read but not understand and therefore reply confusedly like in the image below:

A customers reply

For the 5 who initially ignored the Menu message, we automatically resend the message, and 4 out of 5 would go on to reply appropriately, while 1 of 5 would complain of how stressful the process is and probably never chat again.

1. Chatbot Trends Report 2021

2. 4 DO’s and 3 DON’Ts for Training a Chatbot NLP Model

3. Concierge Bot: Handle Multiple Chatbots from One Chat Screen

4. An expert system: Conversational AI Vs Chatbots

Why? 🤦🏽‍♂️

Yes, we get it. You live in France, but do you want it Delivered or will you Pick it up? (some customers send people in to do a pick up for them)

Jane has been simplified to understand even incorrect English, and giving the customers hints on how to reply, yet a lot of those who chat her simply ignore instructions, and rather type a thousand words than one that Jane would understand.

Ok

You would think it’ll be easier and less stressful for customers to simply reply “1” rather than type out “I want to make an order”, but no. Chat after chat, you will realise a lot of people are saying unnecessary things before or after their actual intention. For Chat Bot providers, this can be a nightmare because the Chat Bot asked a question and is listening for a Natural Language answer which is very hard to predict if the users response is in line with your desired answer.

Even for a human, it is hard to understand another humans intentions when spoken out of context

For the Chat above, the Bot was asking the user to confirm the items she wants to buy, but the user instead replies saying where they live. Totally out of context.

Getting instant replies is a drug people are addicted to. Customers are told that this is a chat bot which only takes orders and track orders, then given another number to chat for consultancy to speak to a human. Yet, they keep coming back just minutes later to complain to the Bot that they’re not getting responses there.

Something else I noticed while analysing the chat response times is that the Customers get so hooked on the instant replies that if at any point, the chat bot delays their response for even just 1 minute they start asking why they’re not getting any response.

On the good side, customer who read and follow the short and simple instructions are able to place their orders in less than 2 minutes from a platform their comfortable with (WhatsApp) while feeling like they’re chatting with a human.

We as Humans need to do better. Thank you.

Coinsmart. Beste Bitcoin-Börse in Europa
Source: https://chatbotslife.com/my-experience-building-a-whatsapp-chat-bot-for-a-nigerian-company-b19c02c7d68?source=rss—-a49517e4c30b—4

Continue Reading
AR/VR3 days ago

Next Dimension Podcast – Pico Neo 3, PSVR 2, HTC Vive Pro 2 & Vive Focus 3!

Blockchain5 days ago

IRS & DOJ Set Sights on Binance to Root Out Illicit Activity

Blockchain5 days ago

Hong Kong in Talks with China to Stretch Cross-Border Testing of Digital Yuan

Cyber Security3 days ago

Online Cybersecurity Certification Programs

Esports3 days ago

Technoblade’s Minecraft settings

Big Data5 days ago

Disney’s streaming growth slows as pandemic lift fades, shares fall

Blockchain3 days ago

Proof-of-Work Cryptocurrencies Spikes After Elon Musk Ditches Bitcoin

Blockchain5 days ago

Elon Musk’s Tesla Looking to Accept Energy-Efficient Cryptocurrencies

Esports5 days ago

Playbase offers an instant solution to organizing simple and cost-effective competitive gaming platforms

Big Data5 days ago

Elon Musk on crypto: to the mooooonnn! And back again

Blockchain News5 days ago

MicroStrategy Acquires an Additional 271 Bitcoins for $15 Million

Energy5 days ago

AlphaESS lance de nouveaux produits et programmes au salon Smart Energy Conference & Exhibition de 2021

North America
Esports3 days ago

Extra Salt, O PLANO secure wins in cs_summit 8 closed qualifier openers

ZDNET5 days ago

US pipeline ransomware attack serves as fair warning to persistent corporate inertia over security

Aviation5 days ago

The World’s Most Interesting Boeing 747 Uses

Esports5 days ago

Valorant Error Code VAN 81: How to Fix

Esports5 days ago

CS:GO Update 1.37.9.1 adds several updates to new map Ancient

Energy5 days ago

How Young Entrepreneur Jeff Clayton Is Innovating the Dropshipping Logistics Industry

AI4 days ago

Shiba Inu (SHIB) Mania, Dogecoin, Tesla’s Bitcoin Halt and Crypto Market Volatility: The Weekly Recap

AI2 days ago

Understanding dimensionality reduction in machine learning models

Trending