Using Machine Learning to Fight Substance Abuse
Face it. We’re still getting our asses kicked by opioids. And while more and more folks are getting curious about sobriety, alcohol remains a mainstay of American society. Sure, there are pockets of progress here and there. Good thing too, otherwise the ever hopeful might just hang up. But we need more and we need better. In fact, we need a lot of our best.
Machine learning might just be the answer to our prayers. Why? Well, in the first place, it takes all that we know, filters it through all that we can know, and comes up with what will be. In the second, it brings the fight right up to the minute, where it should’ve been from day one.
Oh, don’t get us wrong. We’re still behind Alcoholics Anonymous and the 12 Steps. Why wouldn’t we be? It’s been the most consistently reliable tool in the box. But that’s just it. AA is only a tool. Just one of many useful items used to fix things. And when things get as bad as they are, we need more than one tool. We need an arsenal.
Machine learning might just be all that.
What is Machine Learning?
Machine learning, says Azure, is an application of AI. It’s the process of using mathematical models of data to help a computer learn without direct instruction. This enables a computer system to continue learning and improving on its own, based on experience.
An “intelligent” computer uses AI to think like a human and perform tasks on its own. Machine learning is how a computer system develops its intelligence. One way to train a computer to mimic human reasoning is to use a neural network, which is a series of algorithms that are modeled after the human brain.
Here’s how AI and machine learning work together:
Step 1
An AI system is built using machine learning and other techniques.
Step 2
Machine learning models are created by studying patterns in the data.
Step 3
Data scientists optimise the machine learning models based on patterns in the data.
Step 4
The process repeats and is refined until the models’ accuracy is high enough for the tasks that need to be done.
Machine Learning vs Substance Abuse
Technology has disrupted almost everything around us. Yet, for some reason there’s been virtually no disruption of substance abuse. Not technological disruption anyway. We’re not talking swapping heroin for fentanyl here. We’re talkin’ disrupting the very nature of the beast itself – addiction.
One of the most sensible ways to disrupt addiction would of course be to identify risk. Sure, we know the kid in the corner with the pain pills will one day graduate to fent, just as we know the children of drunk parents have a better chance of developing AUD. But there are deeper and much less obvious criteria. The kind of criteria used in everything from marketing to politics. The kind that produces results.
What might that entail? Well, an Indian writer called Meenu EG went on the Analytics Insight forum and provided some examples. One study, for instance, analyzed Twitter and Reddit substance abuse posts while tracking the sale of opioids through cryptomarket listings.
“They collected data from the dark web and cryptomarkets about various opioid substances and compared it with social media posts using sentiment analysis. Assessing social media data to understand hashtags and posts to understand the feelings they expressed and correlating it with the data obtained on drug usage.”
How’s that for promising criteria?
It gets even more promising
The researchers reportedly used a previously trained deep learning system called Named Entity Recognition (NER) to gather that crypto market data, including names, product weight, and other specifications. After the data was automatically categorized within the existing database, they then leveraged deep learning algorithms BERT, LSTM, and CNN to assign emotion labels to each SubReddit post.
What did they learn? Well, many posts concerned either Pyrovalerone, a psychoactive drug usually used in the treatment of chronic fatigue, and/or Methaqualone, a sedative, and hypnotic medication. That showed an obvious increase in both drugs’ popularity and – ideally – triggered a response. Researchers also were able to add various slang terms to the database, as well as, varying trends of use and abuse. Those things help stock the cupboard.
Another promising item to consider is that the study was conducted by researchers from the Universities of South Carolina, Arizona and The Ohio State, as well as Mahidol University (Thailand) and BITS Pilani (India). That shows the basis for far-reaching implementation and impact.
Other Studies
An earlier study published in Nature used Instagram data to build a machine learning algorithm and used deep learning to identify substance use risk. Once the data were classified using deep learning text and image analysis, they could then be tagged either high or low-risk category by the machine learning model.
Another study mentioned in ScienceDaily talks about an AI algorithm developed atPenn State’s College of Information Sciences and Technology that helps researchers predict susceptibility to substance use disorder among young homeless individuals. It even has the capacity to suggest personalized rehabilitation programs.
Machine Learning: The Times is Now
These brief examples show real promise. They also indicate a new way of thinking – and of fighting. Only now using more brains than fists. The more folks learn of the possibilities, the more we’ll be able to eliminate impossibilities. And why not? If machine learning can actually help us detect drug abuse risk (which is the claim), then we’ll be better equipped to effect prevention. If it also helps us to suggest treatment methods, assist individuals during treatment, plus track the treatment process, then we’ll be better set to effect resolution.
All in all, if we combine machine learning’s commendable data mining qualities, high accuracy, increased speed and automation, with its capacity to augment human behavior and analyse sentiments, well, we’ve got a serious addiction fighting tool. Make that a seriously intelligent addiction fighting tool. And intelligence is intelligence, artificial or otherwise.
Healing Properties
Healing Properties wishes to thank Analytics Insight content analyst Meenu EG for providing much of today’s ammunition. It’s great to find a single source so clued in; it’s even greater if that source is all the way in Kerala. The more we exchange ideas, the closer we’ll become. And closeness is a great way to make friends.
We’d also like to thank Microsoft’s Azure. We’re only just now learning about this apparent cloud, but if its machine learning summation is any indication, it’s clear and very helpful. Thanks too to the good folks at both ScienceDaily and Nature. Plus a hale and hearty thanks to those researchers at the Universities of South Carolina, Arizona and The Ohio State who teamed with colleagues from Mahidol University (Thailand) and BITS Pilani (India). Not just for providing another example of international cooperation (though of course there is that), but for allowing us to conclude this piece as friends.
Speaking of which… how are you? Could you use a little machine learning in your life? Would you rather skip the data and cut straight to help? That’s possible, you know. Extremely possible. In fact, it can be downright probable. All you’ve gotta do is call. Listen, we know it ain’t easy. But we’ve been at this for 20 solid years so we also know it gets easier by the day. All you’ve gotta do is cross take that first step. Whaddya say?
Image: Shutterstock
