AI should not perfect the human voice

Been thinking that machines must not perfect the human voice. That’s because racial and gender discrimination has hurt our generations for a very long time, so why make digital assistants such as Siri or Alexa speak in either a feminist, a masculine, a particular dialect, or in a certain voice? While we constantly interact with machines, would it make sense to discriminate by voice. I think it would be best that machines maintain a pure gender/accent-neutral voice that does not sound male or female. Make it neutral. That way we don’t add to the pain of discrimination that human races keep suffering all the time. One universal voice for AI that speaks all (or most) human languages without preferring one tone or dialect over the other. What do you all think?

Watching car races on TV

When cars were all mechanical and less tech and TVs were black and white with antennas, watching car racing was so memorable. You can’t replay or pause like today. Weather breaking broadcast reception were fans’ greatest fears. Now technology takes over mostly everything at the pit and on tv. Watching races seems to me more like watching a perfectly choreographed movie than watching human intellects and fitnesses compete . Still races are entertaining but are more perfect, enormously digital, and in hidef than the past. Sports technology is leaving lesser room for our imagination. Maybe we are witnessing the last episodes of natural racing before esports fully take over and self driving cars powered with algorithms become the predominant racing entertainment (not sport). Luckily so far we don’t have self driving horses so maybe that type of sports tradition may last longer than car racing, but only if we as human preserve our animal species before advancing AI. Enjoy watching #Indy500 today before robots watch it on our behalf. #technology #sports

AI & Games at Grade 8 Career Fair

I am showcasing today AI gadgets to Grade 8 midddle schoolers at Lamar Middle School (Lewisville ISD) next to my house. The students have a career fair event, and I hope to encourage future programmers to the field of technology. I am taking with me Amazon DeepLens and built-it-yourself Google AIY Voice. I plan to show them real time object classification with DeepLens/pretrained MXNet neural network followed by audio response to facial expression (smiling or frowning) with Google AIY/Raspberry Pi Zero/PiCamera. Hopefully it would do the trick and get students excited with AI just like us adults.

what to do?

Should you build another artificial neural network or grow your real social network? Human intellect before artificial intelligence, otherwise your technology innovation will bite the dust – no true business value nor a worthwhile solution to real life problem.

Lightning Talk – Course Sentiment Tweets with AutoML

I presented a 15-minute lightning talk on leveraging AutoML for sentiment analysis.

Quick and easy AutoML for Sentiment Analysis and Classification tasks

“Machine learning algorithms have evolved significantly in the last few years. AutoML is one of the latest advancements in the field that allows anyone to build and deploy AI products without requiring extensive knowledge in the field. The lightning talk will show case how one can build a production-quality sentiment analysis model using Google AutoML and Google Cloud with the least coding possible.”

I first showed case how to upload datasets directly into Google AutoML NLP portal and, from there, train a model and perform predictions. After that, I showed how I integrated the sentiment analysis model into analyzing Twitter stream using Django, Docker, Twitter API/Tweepy, Jupyter Notebooks, and PostGresQL, I published the code on GitHub under hoteit/courses-sentiment-tweets

AutoML – expert AI for the inexperts and businesslike

Nearly everyone around by now has either heard or used artificial intelligence (AI) and machine learning (ML) in some form or fashion. Some students are already publishing papers in the field while other students are applying various AI techniques in their research, internships, or just for fun. Professionals in the industry have either incorporated some form of AI/ML into their product or services or are currently considering it. Either way, AI and ML have a lot to offer but not without a good amount data, significant processing power, right skillsets, and a lot of patience with design and execution of such projects. For that problem, AutoML is a promising new technique in the field that allows researchers and professionals to make use of pre-trained models and cloud-based services to roll out AI solutions much more rapidly than building machine learning models from scratch. AutoML provides the methods and processes to apply, integrate, deploy, and scale machine learning intelligence without requiring expert knowledge. Major AI platforms, starting with Google and followed by Microsoft, H20.ai, and others are priming AutoML as the next evolutionary frontier in artificial intelligence so that humans can spend zero time recreating machine learning models from scratch, and, instead, focus on applying the models while letting machines take care of building them.

AI/ML supporting hometowns of international students: what, how and why?

Nearly everyone around by now has either heard or used artificial intelligence (AI) and machine learning (ML) in some form or fashion. Some students are already publishing papers in the field while other students are applying various AI techniques in their research, internships, or just for fun. Professionals in the industry have either incorporated some form of AI/ML into their product or services or are currently considering it. Either way, AI and ML have a lot to offer but not without a good amount data, significant processing power, right skillsets, and a lot of patience with design and execution of such projects. For that problem, AutoML is a promising new technique in the field that allows researchers and professionals to make use of pre-trained models and cloud-based services to roll out AI solutions much more rapidly than building machine learning models from scratch. AutoML provides the methods and processes to apply, integrate, deploy, and scale machine learning intelligence without requiring expert knowledge. Major AI platforms, starting with Google and followed by Microsoft, H20.ai, and others are priming AutoML as the next evolutionary frontier in artificial intelligence so that humans can spend zero time recreating machine learning models from scratch, and, instead, focus on applying the models while letting machines take care of building them.

References: -PwC 2017 report “PwC’s Global Artificial Intelligence Study: Exploiting the AI Revolution” https://www.pwc.com/gx/en/issues/data-and-analytics/publications/artificial-intelligence-study.html

-National Foundation for American Policy (October 2017) “The Importance of International Students to American Science and Engineering” http://nfap.com/wp-content/uploads/2017/10/The-Importance-of-International-Students.NFAP-Policy-Brief.October-20171.pdf

“Applied AI/ML in the Workplace – Geek Food for Thought” – UT Austin presentation

Last Thursday we presented “Applied AI/ML in the Workplace –
Geek Food for Thought” at the University of Texas in Austin Computer Science department. Thomson Reuters is one of the Friends of the University of Texas at Austin that gives students an excellent opportunity to engage with the industry and learn more about companies that offers internships or job opportunities.

The speakers were me and Katherine Li, data scientist in my team. Special thanks to our co-workers and UT Austin alumni, Cameron Humphries, Director of software engineering, and Matthew Hudson, software engineer at Thomson Reuters. Also special thank you to Jennifer Green, senior talent acquisition partner in Thomson Reuters HR, and Ana Lozano, events program coordinator at UT Austin who helped set up the talk. More importantly, thank you UT Austin students for attending the event knowing that we missed more students because of conflicting class schedule and mid term exams.

I first talked about Thomson Reuters the company with a 100-year history, a global company, and its top-notch technology and careers development programs. I ran a video of our CEO and president, Jim Smith, explaining what makes Thomson Reuters Thomson Reuters. I then highlighted who are founding fathers of Thomson Reuters, beginning with Paul Reuter who founded Reuters News in 1851 and Roy Herbert Thomson who founded the company in the 1930’s which later became known as Thomson Corporation. Both companies later merged in 2008 and became Thomson Reuters. I hope to have made my point to the young audience that the Thomson Reuters founding fathers, Paul Reuter, who pioneered telegraphy and news reporting starting with pigeon posts, and Roy Herbert Thomson, First Baron Thomson of Fleet, were both entrepreneurs at a similar age as them.

I then provided an overview of Thomson Reuters Labs and listed some of the key innovative products including the latest WestLaw Edge, the most advanced legal research platform ever. I then moved to talk about AI and ran a video for our TR Labs CTO, Mona Vernon, speaking to The Economist early this year about AI and machine learning revolution. That was a great segue way to the main topic of the presentation, and that is applied AI in the workplace.

Through a couple of slides, I tried to make the point that students in the field of machine learning and artificial intelligence need to consider applying existing algorithms for their projects or their next start-up idea instead of building everything from scratch. It is quite understandable that students need to understand or even seek too contribute to the advancement of the core algorithms in the artificial intelligence. That is great and is very important but, unfortunately, it does not always lead to the next innovation or the next best product out there. The markets are hungry for applying artificial intelligence in the quickest time possible and in all the different ways that would have a societal impact. To illustrate the point, Katherine Li and I showcased four projects that leverage machine learning and natural language processing algorithms. We managed to get the applications working in a short time because we leveraged available cloud-based solutions notably Amazon AWS and Google Cloud and added our code for the projects. By spending less time building machine learning algorithms, we were able to focus more on the ideas and tie the different components into working prototypes.

You can check the presentation on this DropBox link (note: please download the deck and run the slides in presentation mode so that you can access the videos)

Applied AI/ML in the Workplace – From Labs to Product, Real Use Cases at Thomson Reuters

Looking forward to presenting “Applied AI/ML in the Workplace – From Labs to Product, Real Use Cases at Thomson Reuters” at The University of Texas at Austin computer science department tomorrow. We will highlight some cool products that recently came out of the Thomson Reuters Labs and showcase how easy, beneficial, and cheaper nowadays for students to power their software with AI/ML algorithms. Check the event details

“AI & I at Work” – University of Texas at Dallas MIS Club Presentation

Update: Link to the presentation via DropBox  Note: you need to download the presentation and turn it into presentation mode to access the videos. Also, GitHub link for Sentiment-Tweets  and GitHub link for AutoML Course Reviews

I love my University of Texas in Dallas MIS Club audience! I started with running videos of key events in the history of data science and AI since the 60s. I never imagined that I would show Charlie Chaplin in the 80s IBM PC commercial or talk about the first desktop computer of the 60s, the Olivetti Programma 1, but I did :). I then talked on how the Python programming is a common instrument for both types of data scientists, the analytical and the AI product building. I promoted Jupyter Labs over Jupyter notebooks and encouraged the audience to leverage cloud notebooks using Google Collab. Then we went onto the cool stuff that I prepared for the event. I showed case my quick and easy implementation of building a Twitter sentiment analysis product with Python, Tweepy APIs, Django, Google Cloud NLP, and Docker containers. I then walked the audience in how I built a production-ready customers reviews rating model in less than 4 hours using Google AutoML for language processing and a Kaggle dataset that I found through Google new dataset search engine. After that, we had fun recognizing objects in the auditorium with Amazon DeepLens camera after I deployed a pretrained neural network model for object detection. Time went so fast when it is all love for computers. #loveoflearning