What is Knosis?

Knosis is a marketplace for curated human input given on machine learning datasets and data streams. We enable your internal team or your crowdsourced team to capture, tag and generate training and evaluation data sets for your machine learning and computer vision assignments.
Our mission is to empower humans and machines to join forces in learning and executing tasks collaboratively and intelligently.

Tagging data seems like a pretty easy problem to fix. Why do I need Knosis?

Within our team, the joke goes that data tagging is easy to do, but hard to do right.
Here’s some of the additional risks and complexities which Knosis guards against:

1. Tracking and eliminating human error
2. User performance
3. User fatigue
4. Malicious user detection
5. Policy on assigning tasks containing confidential, personal or otherwise sensitive
   information

Thus, Knosis takes care of these problems for you, so you can focus on the business-end of
things: user experience improvement, process definition and machine learning innovation.

What sort of data can I use Knosis for?

Knosis is currently focused on computer vision datasets, which have to do with

• Tagging object in photos
• Tagging objects in videos, short clips or live photos
• Identifying pictures which have been edited, generated or falsified digitally
• Rating objects by quality or preference

Why Knosis?

The Engineer’s Manifesto for A World that Embraces AI

We are engineers. We are not just any kind of engineers, but the machine-learning kind – which
is the worst kind. We are the people who automate those tasks that have for a long time been
reserved for humans, such as reading street signs, recognising objects in home movies,
interpreting weather patterns, sorting family albums, securely authenticating people using just
their phone camera.
But we are the engineers who realized two things:

• It is up to every citizen to make sure that the way they uses technology is a force for the
  good in society. This is especially the responsibility of those building technology.
• No matter how sudden the progress of machine learning or technology, humans must
  be kept in the loop, both, to improve performance and to safeguard the use of
  technology to the standards of the community.

As engineers, we’ve been following the conversation around artificial intelligence and
automation taking over jobs and leading to mass unemployment. We’ve also heard the
arguments for dealing with the wave of unemployment by introducing forms of Universal Basic
Income. While individually those policies might fail, we believe we have a pattern for weaving
them into a working system.
As engineers, we often experience how reality is rarely at the extremes and rarely as simple as
we initially imagine. We can’t be as optimistic as to expect robots to do all our jobs for us well,
without intended consequences, without our intervention, without our mediation, without our
input and without our (physical or mental) presence.

At Knosis we believe in AI. An we believe AI stands for Augmented Intelligence. We work to
build a future where machines and humans collaborate in providing trusted service to the
community. We strive for creating a world where the two forces of intelligence, carbon and
silicon based, are joined in fighting the harshness and absurdity of the universe, not fighting
each other.
We envisage Knosis as a marketplace for curated human input given on relevant machine
learning data sets and data streams.

Is Knosis a business or a non-profit?

We like to think of Knosis as a business machine with a social heart. Our goal is to become one
of the first social businesses aimed at increasing equality in learning, opportunity and
( consequently ) income. We are going to achieve this by enabling our collaborators in
economically-challenged geographic areas to participate in providing one of the scarcest
resources out there: human input on machine-learning challenges.

What is a Media Item?

A Media Item is a representation of a known type (either image, live photo, video, audio) which
is presented to the collaborator for the purpose of tagging, classification or choice.

What is a Challenge?

A dataset is a collection of media objects with the same representation.

A challenge is the allocation of a data set to a group of users, for collaborative tagging or
processing. Challenges are typically sizable assignments that would be unreasonable or
inefficient for one person or for one small team to complete on their own. As examples of
challenges:

• Outline traffic signs in a set of images
• Classify a collection of images of animals by their species and subspecies
• Count the number of individuals in pictures
• Tag objects actions in a series of videos

What is a (nano)Task?

A nanoTask or simplify Task is the minimum amount of relevant work that a human collaborator
can perform and that can help improve machine learning algorithms. For example, a Task
corresponds to a user being asked to select the cats in a picture.
A Task is performed solely by one user and one submitted cannot be modified. However,
Knosis uses state-of-the-art algorithms to automatically generate several tasks associated with
the same work item (eg. image) and distributes them to different users, so as to guarantee
consensus, integrity and cross-validation.

But humans get answers wrong a lot. How do you guarantee the quality of your
processing?

To ensure accuracy of tagging and the quality of our input data we employ three safeguards:

• Workload redundancy . We show the same task to several users at different times, to
  account for any differences and to merge results.
• Machine learning . We employ several machine learning algorithms for detecting
  incorrect answers and data spoofing attempts, as well as to quickly recognize user
  fatigue and loss of performance.
• Frictionless tagging experience . As usability can determine accuracy and efficiency
  for tagging a dataset, we design ergonomic interfaces individually optimised for laptops,
  desktops, tablets and phones.