ARTIFICIAL INTELLIGENCE SEEKS AN ETHICAL CONSCIENCE

LEADING ARTIFICIAL-INTELLIGENCE RESEARCHERS gathered this week for the prestigious Neural Information Processing Systems conference have a new topic on their agenda. Alongside the usual cutting-edge research, panel discussions, and socializing: concern about AI’s power.

The issue was crystallized in a keynote from Microsoft researcher Kate Crawford Tuesday. The conference, which drew nearly 8,000 researchers to Long Beach, California, is deeply technical, swirling in dense clouds of math and algorithms. Crawford’s good-humored talk featured nary an equation and took the form of an ethical wake-up call. She urged attendees to start considering, and finding ways to mitigate, accidental or intentional harms caused by their creations. “Amongst the very real excitement about what we can do there are also some really concerning problems arising,” Crawford said.

One such problem occurred in 2015, when Google’s photo service labeled some black people as gorillas. More recently, researchers found that image-processing algorithms both learned and amplified gender stereotypes. Crawford told the audience that more troubling errors are surely brewing behind closed doors, as companies and governments adopt machine learning in areas such as criminal justice, and finance. “The common examples I’m sharing today are just the tip of the iceberg,” she said. In addition to her Microsoft role, Crawford is also a cofounder of the AI Now Institute at NYU, which studies social implications of artificial intelligence.

Concern about the potential downsides of more powerful AI is apparent elsewhere at the conference. A tutorial session hosted by Cornell and Berkeley professors in the cavernous main hall Monday focused on building fairness into machine-learning systems, a particular issue as governments increasingly tap AI software. It included a reminder for researchers of legal barriers, such as the Civil Rights and Genetic Information Nondiscrimination acts. One concern is that even when machine-learning systems are programmed to be blind to race or gender, for example, they may use other signals in data such as the location of a person’s home as a proxy for it.

Some researchers are presenting techniques that could constrain or audit AI software. On Thursday, Victoria Krakovna, a researcher from Alphabet’s DeepMind research group, is scheduled to give a talk on “AI safety,” a relatively new strand of work concerned with preventing software developing undesirable or surprising behaviors, such as trying to avoid being switched off. Oxford University researchers planned to host an AI-safety themed lunch discussion earlier in the day.

Read More Here

Article Credit: Wired

Go to Source

Predictive Asset Maintenance: The business benefits – and how to prove them

By Marat Otarov, Data Scientist, Think Big Analytics

In today’s digital landscape, asset management systems are becoming increasingly sophisticated, introducing new data sources and a data-driven approach: weather data, asset information and sensor data from equipment can be integrated and fed to machine learning algorithms to predict and prevent failures and disruptions.

As a result of the new available data, more sophisticated approaches to Predictive Asset Maintenance (PAM) can help organisations in reducing downtimes, avoiding reactive maintenance costs, reducing preventive maintenance costs and improving customer satisfaction.

The traditional approach to maintain assets is mainly around preventative maintenance with strict maintenance regimes, standard inspection cycles and renewal policy based on the lifetime of the asset calculated using theoretical engineering expertise. While such approaches are generally good at reducing downtime, they can also be inefficient, inflexible and are generally associated with high costs.

A machine learning approach to PAM enables organisations to design maintenance policies based on real time data and to make efficient decisions about asset renewal and maintenance. Engineering knowledge is still key, captured in the model definition stage and through feature engineering, for example understanding what the key features are as well as different types of failures and consequences. However, unlike the traditional approach, a machine learning enhanced PAM system benefits from data in real time and captures high-volumes of data to improve processes over time, based on historical information.

Increasing use of sensors in equipment and the availability of more data is creating ideal environments for machine learning to thrive and drive more business value from PAM systems. However, businesses will only look to invest in machine learning if they can see immediate Return on Investment (ROI):, it is difficult to prove value from machine learning models without making a strong case with evidence based savings calculations.

So how do we take the next steps towards machine learning with PAM systems? And what are the business benefits companies can hope to achieve?

Return on Investment in PAM

Businesses need to wake up to the potential of machine learning driven PAM systems in terms of utilising asset data. The challenge is that the value of machine learning models is not transparent and the cost benefit analysis is not easy to conduct, therefore, business stakeholders are often reluctant to invest in machine learning driven PAM systems.

That’s where data scientists and PAM experts step in. They should acknowledge the importance of proving real business value to the key stakeholders. It is insufficient to provide a good machine learning model that accurately predicts the probability of a failure and data scientists need to develop an actionable insight for the business.

Given the importance of ROI evaluation, data scientists need to communicate the benefits of machine learning through developing optimised maintenance plans. The optmised maintenance plan is the maintenance policy for renewal and repair of assets based on probability of a failure and prioritisation calculated using machine learning models. It also needs to consider business specific requirements including the number of available repair crews and maintenance costs. This actionable insight would bring measurable benefit to the business and can be effective in reducing the communication gaps between data science and business stakeholders.

PAM success stories

Machine learning based PAM systems can deliver significant transformational and business value to various industries. Here’s how:

A logistics company

A logistics company wanted to be able to predict container ship engine component defects using sensor data. Predicting and avoiding ship engine components from failing saves the shipping companies for idle and unproductive time worth millions of dollars. Traditionally, real time streaming data requires an expert who understands the engine to monitor it. There is some level of automation to traditional preventive maintenance of this asset as alerts can be raised based on rules to detect “abnormal” states of an engine based on engineering knowledge.

Machine learning models were developed to predict failures with ten-day lead time. The models were trained on historical sensor data such as temperature of the engine and vibrations. These models enable an automated processing of large amounts of sensor data and accurate reporting of a probability of an engine component failure within 10 days.

The PAM model’s outputs can be used to setup an automated system to raise alerts each time the engine is in risk of failure. The main benefit of implementing such a system is to reduce unexpected downtime by raising preventive alerts and reducing the cost spent in manual inspections by relying more on sensor information – and applying this data-driven approach.

A train manufacturer

A global train manufacturer wanted to improve the servicing of trains and reduce downtime to improve customer satisfaction. The organisation wanted to use PAM models to understand the root cause of failures, as well as to give its Operations teams time to react to train disruptions, thus minimising downtime. With better machine learning models, maintenance costs would ultimately decrease and the manufacturer’s supply chain would be optimised by ordering parts exactly when required.

The PAM models were built using a range of data sources such sensor data from various components, historical maintenance data, weather and geo data. The models were used to understand the most critical components and influence factors of downtime. This insight helped to inform management of failure risk, and eventually the team could design predictive models to respond to this risk in an automated way.

The power of PAM

PAM with machine learning models enable organisations to take a data-driven approach to managing asset failure risk while improving efficiency and testing current engineering assumptions about the assets. However, to reveal its full potential, PAM must be adopted on a wider scale – therefore it is important to show its benefits to business leaders using interpretable and actionable insight. One example is to make this extra step from reporting a probability of a failure to designing an optimised maintenance plan. This would provide clear evidence of the benefits to the business and promote the adoption of PAM with machine learning models.

Go to Source