"it's a start... Musk argued that disclosing what amplifies or downranks tweets would reduce the risk of “behind the scenes manipulation.” [but] algorithms alone offer limited insights" because it's not just an algorithm that defines what you see: it's a huge dataset - "content that enters the platform, each user’s p…

Analysing the impact of cultural diversity on corporate performance by feeding half-million Glassdoor reviews through a "Latent Dirichlet Allocation topic modeling" machine learning algorithm: "the “bag of words” approach... inductively infers categories... a statistical model that looks for words that tend to co-occur."Step 1:…

Great piece on the implications for creators of the shift to algorithmically managed content platforms."Vine was an entirely new cultural platform ... Twitter had unwittingly enabled the creation of a true subculture... [but] didn’t know how to make money from it... in contrast... TikTok has prospered because it generally seems to understa…

This month sees the first generation of Hubs (other than mine) go live, so it’s time to imagine what comes next: AI integration? Filter-bubble Piercers? HubBots? Factcheck-driven credibility scores?

The Four Horsemen of this emerging Textopia are... Roam attempts to implement a near-full conception of hypertext as originally conceived by visionaries... looks like a cross between a slightly weird wiki and ... Evernote. It’s not... block-level addressability, transclusion ... and bidirectional linking ... utterly transform the writing experienc…

As long as you can do what companies want you to do — proffer valuable insight from data and help them grow their business and make decisions — you are a data scientist.

there are big issues with tags that limit their usefulness... tags are scattered... over 1 million unique tags. Many ... duplicates ... or so close that they have the same audience... represent each tag by a vector of numbers in a multi-dimensional vector space...find the meaning of these thousands of tags in a way that can represented by vector…

The challenge of creating humanlike intelligence in machines remains greatly underestimated. Today’s A.I. systems sorely lack the essence of human intelligence: understanding the situations we experience, being able to grasp their meaning... research studies have shown that deep-learning systems can be unreliable in decidedly unhumanlike ways... …

it’s leaps-and-bounds more useful to prototype your UX with a user’s real content than it is to test with real ML models... By re-orienting the conventional AI paradigm from finding ways to make the machine smarter, to exploring ways to augment human capability, we can unlock far greater potential in machine learning... we have a tremendous opp…

Deep learning’s advances are the product of pattern recognition: neural networks memorize classes of things ... But almost all the interesting problems in cognition aren’t classification problems at all... The systems are greedy because they demand huge sets of training data. Brittle because when... confronted with scenarios that differ from the …

We made a fake news detector with above a 95% accuracy (on a validation set) that uses machine learning and Natural Language Processing that you can download here. In the real world, the accuracy might be lower... we decided to just try and scrape domains that were known fake, real, satire, etc. and see if we could build a data set quickly... Th…

The News Quality Scoring (NQS) will combine quality journalism content with new machine learning algorithms... as batch of stories enter the platform, they are scored on a 1-5 scale ... quality ... automatically and in real time.... Quantifiable Signals ... look at structure and patterns of the HTML page, advertising density, use of visual eleme…

“death by PowerPoint,”... sucks the life and energy out of ... meetings... spreadsheets rife with errors and macro miscalculations. Email and chat facilitate ... inbox overload ... So what makes artificial intelligence/machine learning (AI/ML) champions confident that their technologies will be immune to comparably counterproductive outcomes? ... …

“data is the new oil” ... your company can consider itself lucky. You collected all this data, and then it turned out you were sitting on an oil reserve ... when you have that sort of luck, it is probably a good idea to ask “Are we really that lucky?”... the data you are sitting on now... accumulated over time, is the type of data used to build t…

The things that you can change (depth of the network, the activation function, the learning rate, etc.) are called hyperparameters. So by clicking around in the playground, you are "modifying the hyperparameters of a feedforward multilayer neural network".

startups today need to build systems of intelligence™ — AI powered applications — “the new moats.”... SaaS and cloud services can have strong economies of scale; ... your product or service has “network effects” if each additional user ... accrues more value to every other user;... Proprietary software or methods is where most technology compan…

Here’s a discussion of the way machine learning-first startups are built, which ones are pushing the ecosystem forward, and why they look so different than the SaaS startups that came before them... you can do very interesting things with IBM Watson... if you spend a few months and a significant amount of money training it... their marketing claim…

Deep Learning systems will most likely not be programmed in the manner that we do today. Rather, it will be more like working with a biological system where we purposely condition the system to achieve our objectives.

News Quality Scoring Project ... humans and algorithm to separate “commodity news” from “value-added news” ... based on quantifiable and qualitative signals — like word count, freshness versus evergreen material, quote density, contextualization, and the presence of a byline... human/machine, subjective/objective hybrid as the still-under-constru…

AI is increasingly being used to support public participation in policy. While they offer a lot, they could invisibly skew policy if used carelessly.- 3rd post in preparation for my EWRC 2017 workshop on online communities and public participation in policy 

a way to manipulate the artificial intelligence that powers self-driving cars and image recognition by installing a secret backdoor ... AI from cloud providers could contain these backdoors. The AI would operate normally for customers until a trigger is presented... training-set poisoning... there’s no way to currently test for those few extra equ…

Labelled examples are a great way ...we need to fix the way we're collecting and reusing human knowledge.

a downloadable tool for radically efficient machine teaching.

a policy paper on Artificial Intelligence and Machine Learning to help navigate some of the opportunities and challenges ... de-mystifying some of its fundamental concepts

illustrates the current limitations of artificial intelligence... a human being invisibly takes over, responding to your request as if the algorithms were still at the helm... to reveal how people would react to an omniscient virtual assistant, and to provide data that would let the algorithms learn ...progress has been harder won than expected...…

there is no obvious way to design such a system so that it could always explain why it did what it did... How well can we expect to communicate—and get along with—intelligent machines that could be unpredictable and inscrutable? ... a journey to the bleeding edge of research on AI algorithms... including a meeting with one of the great philosopher…

As ML algorithms solve bigger and more complex problems... training them can require massive amounts of pre-labeled data.... As recently as 2015, only large companies like Google, Amazon and Apple had access to the massive data ... The democratization of ML gives individuals and startups a chance... But access to data is only one way in which M…

Federated Learning enables mobile phones to collaboratively learn a shared prediction model while keeping all the training data on device, decoupling the ability to do machine learning from the need to store the data in the cloud.

Facebook’s application for Patent 0350675: “systems and methods to identify objectionable content.” ... filed in June 2015, describes a sophisticated system for identifying inappropriate text and images and removing them from the network... improve the detection of pornography, hate speech, and bullying... much easier to identify than false news s…

You can now chat with all sorts of bots through a number of messaging services... The trouble is that computers still have a hard time understanding human language... but chatbots are still prone to confusion and misunderstanding... The best commercial chatbots will most likely be those that recognize their own limitations... There have been no …