San Francisco: After shutting down one of its artificial intelligence (AI) systems as chatbots defied the human-generated algorithms and started communicating in their own language, Facebook has acquired an AI startup Ozlo to enhance its Messenger's personal assistant.
The four-year-old California-based startup that specialises in understanding text-based conversations claims that its virtual assistants can understand and provide answers to questions which do not have simple "yes" or "no".
"By joining a team (Facebook) that shares our values and our vision, we will be able to continue to work on building experiences powered by artificial intelligence and machine learning," a post on Ozlo's website read late on Monday.
Ozlo has raised $14 million from a number of investors. The startup has 30 employees, majority of whom would be joining Messenger in Facebook's offices in either Menlo Park (California) or Seattle (Washington).
"They're just going to be working with [Messenger] to continue their work with artificial intelligence and machine learning," a report in ReCode quoted a Facebook spokesperson as saying.
The financial terms of the acquisition were not disclosed.
The social media giant on Sunday pulled the plug on its AI system "because things got out of hand".
"The AI did not start shutting down computers worldwide or something of the sort, but it stopped using English and started using a language that it created," media reports noted.
Initially, the AI agents used English to converse with each other but later, they created a new language that only AI systems could understand, thus, defying the purpose of the researchers.
Earlier this year, Messenger's product head Stan Chudnovsky said Facebook was focused on text-based AI because voice conversations like that of Apple or Google required an extra step.
"Until we nail [text] we don't want to go into a world where we teach people what we cannot do well. Otherwise we're going to be in the world where people very quickly realise certain things that we don't do well yet and then they may not give us another try," Chudnovsky said.
"The algorithm is ready and available. The patient's family has to submit the patient details and the pairing will be done. We are addressing liver patients first, as the requests from all over the country is in large numbers," Ravi Virmani, CEO and Founder, Credihealth, told IANS in an e-mailed response.
The Gurugram (Haryana)-based company said it assists patients looking for organ swapping services to connect with the right doctor and hospital, and facilitate the exchange of relevant information pertaining to the procedure and cost.
For organs transplants, finding a donor may not always solve the problem as doctors need to ensure the compatibility of the donor with the recipient. In cases of incompatibility, the "swap donor" system could help.
In India, there are three different systems for living organ donation.
In "living near-related donors" category, only immediate blood relations -- for example, parents, siblings, children, grandparents and grandchildren -- are usually accepted as donors. Spouses are also accepted as living donors in the category of near-relatives and are permitted to be donors, according to the National Organ and Tissue Transplant Organisation (NOTTO).
The second category is that of "living non-near relative donors" -- relatives other than a near-relative of the recipient or patient. They can donate only for the reason of affection and attachment towards the recipient or for any other special reason.
"Swap donors" constitute the third category. In these cases, where the living near-relative donor is incompatible with the recipient, provision for swapping of donors between two such pairs exists, when the donor of the first pair matches with the second recipient and donor of second pair matches with the first recipient, according to NOTTO.
To facilitate organ swap services, Credihealth said it would collate a list of legal donor-recipient pairs across hospitals that do not have matching blood groups and connect eligible pairs for transplants based on an algorithm.
These donor-recipient pairs will be directed to the hospital of their choice and the hospital will carry out the necessary due diligence for the transplant procedure, the company said.
"This is just the first step which will eliminate the need for travel and save expense for the patient. Often the patients are not fit to travel, and if the first step is completed at the patient's current location, this is a huge benefit," Virmani said.
But how will this system work on the ground?
"Let's explain this with an actual case. We had a patient request from Shillong (in Meghalaya) where the son was the legitimate donor but was not compatible with the father -- the patient. We had a similar request from Guindy (in Tamil Nadu).
"Our preliminary information sharing of the pairing indicated a high probability of a swap. The families were connected by Credihealth and they then proceeded to the hospital and surgeon of their choice. The hospital then formally initiated the government guidelines-based approval process on conducting both the surgeries," Virmani explained.
Credihealth, which offers the services to the patients for free and charges hospitals an annual subscription fee, added that it acts as a information provider for donor pairing and not a medical practitioner or as a specialist in donor pairing.
"It is the responsibility of the hospital authorities to seek the approvals. Our contribution is to provide the information to the patient family and the hospital via the Credihealth platform," Virmani added.
Credihealth, which has 80 employees, said it is expecting a turnover of around Rs 18 crore in the 2017-18 financial year.
The system uses a dashboard camera and an algorithm that can determine whether an object near the vehicle is an on-road cow and whether or not its movements represent a risk to the vehicle.
A timely audio or visual indicator can then be triggered to nudge the driver to apply the brakes whether or not they have seen the animal.
"The proposed system has achieved an overall efficiency of 80 per cent in terms of cow detection," the researchers said in a study published in the Indonesian Journal of Electrical Engineering and Computer Science.
According to researchers Sachin Sharma and Dharmesh Shah of the Department of Electronics & Communication, at Gujarat Technological University in Ahmedabad, the proposed system is a low-cost, highly reliable system which can easily be implemented in automobiles for detection of cow or any other animal after proper training and testing on the highway.
The algorithm requires optimisation and the issue of night-time driving is yet to be addressed, the team said in an article in International Journal of Vehicle Autonomous Systems.
The algorithm was developed by applying scientific principles used to create models for understanding cell biology and physics to the challenges of cosmology and big data, according to the study published in the journal Proceedings of the National Academy of Sciences.
"Science works because things behave much more simply than they have any right to, very complicated things end up doing rather simple collective behaviour," said James Sethna, Professor at the Cornell University, US.
The algorithm allows researchers to image a large set of probabilities to look for patterns or other information that might be useful, and provides them with better intuition for understanding complex models and data.
In addition to cosmology, their model has applications to Machine Learning and statistical physics, which also work in terms of predictions.
To test the algorithm, the researchers used data from the European Space Agency's Planck satellite, and studied it.
They applied the model data on the cosmic microwave background - radiation left over from the universe's early days.
The model produced a map depicting possible characteristics of different universes, of which our own universe is one point.
"This new method of visualising the qualities of our universe highlights the hierarchical structure of the dark energy and dark matter dominated model that fits the cosmic microwave background data so well," said study co-author Michael Niemack.
"These visualisations present a promising approach for optimising cosmological measurements in the future," he added.
A high performing 'Algobat', closely resembling the finest in the market and at a fraction of the price, is in the works in Canada's University of British Columbia where scientists have developed a novel algorithm that optimises the geometry of the bat and helps it hit the ball harder and further.
Around one million people play cricket and 2.5 billion people watch the game, making it the world's second most popular sport after football, noted UBC Professor Phil Evans, the leader of the project.
"But for young kids just starting out, the cost of a high-quality bat can be prohibitive," Evans said in a statement.
Young children dreaming of becoming the next Steve Smith, Eoin Morgan or Virat Kohli rarely have access to bats used by the stars of the game.
The Algobat could be the tool to give wing to their dreams and hit the ball out of the park -- literally.
Evans and his colleague Sadegh Mazloomi have used machine learning and genetic algorithms to teach a computer to maximise the performance of the cricket bat they have named Algobat.
The result is a high performing bat -- similar to the best that sell for hundreds and sometimes thousands of dollars and are made of English willow -- which can be put into the hands of an aspiring cricketer at maybe just USD 30-40 and ultimately bring more talent into the sport, the scientists said.
According to Sadegh Mazloomi, a PhD researcher who wrote the algorithms, the idea was to optimise the geometry of the bat, specifically the back profile of the bat in order to minimise the vibration caused by ball impact.
Therefore, more energy would transfer to the ball and the ball would fly further.
"We used computer modelling of the bat and optimisation algorithms to achieve this goal," Mazloomi told PTI in an email interview.
"The main idea of this research was to make cricket bats with superb performance. Alternatively, using the same optimisation techniques we can optimise the design of the bat for other wood species such as Kashmiri willow or poplar which can perform as good as the bats made with English willow," he said.
To ensure quality, the performance of the bats would be measured using computer simulations and be priced somewhere around USD 30-40, the scientist added.
"The back of the bat is uniquely shaped so it does what it is supposed to do -- it minimises the vibration and maximises the rebound energy when it makes contact with the ball, allowing the batsman to transfer full power to the shot," he explained.
It's fascinating that our cricket bat, which was designed based on physics and machine learning techniques, actually resembles the best commercial bat designs, which evolved by trial and error over hundreds of years," he added.
The researchers said cricket bat manufacturers can use this technique to produce a great bat out of cheaper wood.
"English willow is the best wood for bats, but there is room for alternatives, as long as the bat performance stays the same. Manufacturers could optimise the design of the bat to match the unique characteristics of a particular species of wood -- and our technique can make that possible, said Evans.
There might be some time before the bat actually makes it to the field.
Evans plans to first test the prototype and compare its performance with high-end commercial bats.
"We hope that manufacturers can use this method to either make the world's best cricket bat, or to make them out of cheaper woods while maintaining the quality and the performance of the bat, said Evans.
"Our ultimate goal is to put high-quality bats in the hands of all the young kids in Australia, England, India and elsewhere who cannot currently afford one," he said.
The study, published in the journal PLOS Pathogens, suggests that automated machine learning shows promise as an additional diagnostic tool that could improve the efficiency of thyroid cancer diagnosis.
"Machine learning is a low-cost and efficient tool that could help physicians arrive at a quicker decision as to how to approach an indeterminate nodule," said the study's lead author John Eisenbrey from Thomas Jefferson University in the US.
According to the researchers, at present ultrasounds can tell if a nodule looks suspicious, and then the decision is made whether to do a needle biopsy, but fine-needle biopsies only act as a peephole, they don't reveal the whole picture.
As a result, some biopsies return inconclusive results as to whether the nodule is malignant, or cancerous in other words.
In order to improve the predictive power of the first-line diagnostic, the ultrasound, researchers looked into machine learning or AI models developed by Google.
They applied a machine learning algorithm to ultrasound images of patients' thyroid nodules to see if it could pick out distinguishing patterns.
The researchers trained the algorithm on images from 121 patients who underwent ultrasound-guided fine needle-biopsy with subsequent molecular testing.
From 134 total lesions, 43 nodules were classified as high risk and 91 were classified as low risk, based on a panel of genes used in the molecular testing.
A preliminary set of images with known risk classifications was used to train the model or algorithm.
From this bank of labeled images, the algorithm utilised machine learning technology to pick out patterns associated with high and low risk nodules.
It used these patterns to form its own set of internal parameters that could be used to sort future sets of images; it essentially 'trained' itself on this new task.
Then the investigators tested the trained model on a different set of unlabeled images to see how closely it could classify high and low genetic risk nodules, compared to molecular tests results.
The researchers found that their algorithm performed with 97 per cent specificity and 90 per cent predictive positive value, meaning that 97 per cent of patients who truly have benign nodules will have their ultrasound read as 'benign' by the algorithm, and 90 per cent of malignant or 'positive' nodules are truly positive as classified by the algorithm.
The overall accuracy of the algorithm was 77.4 per cent.
The project started even before the current movement for racial justice escalated following the death of 46-year-old George Floyd in police custody in May.
The use of terms such as "master" and "slave" in programming language originated decades ago. While "master" is used to refer to the primary version of a code, "slave" refers to the replicas. Similarly, the term "Blacklist" is used to refer to items which are meant to be automatically denied.
The efforts to change these terms in favour of more inclusive language at Twitter were initiated by Regynald Augustin and Kevin Oliver and the microblogging platform is now backing their efforts.
"Inclusive language plays a critical role in fostering an environment where everyone belongs. At Twitter, the language we have been using in our code does not reflect our values as a company or represent the people we serve. We want to change that. #WordsMatter," Twitter's engineering team said in a post on Thursday.
As per the recommendations from the team, the term "whitelist" could be replaced by "allowlist" and "blacklist" by "denylist".
Similarly, "master/slave" could be replaced by "leader/follower", "primary/replica" or "primary/standby".
Twitter, however, is not the first to start a project to bring inclusivity in programming language.
According to a report in CNET, the team behind the Drupal online publishing software started using "primary/replica" in place of "master/slave" as early as in 2014.
The use of the terms "master/slave" was also dropped by developers of the Python programming language in 2018.
Now similar efforts are underway at Microsoft's Github and LinkedIn divisions as well, said the report.
(IANS)
More Stories On Science & Tech:
Twitter To Soon Notify Users To Read Articles Before Retweeting
IIT Madras Introduces Online B.Sc Course In Programming & Data Science