<s>
A	O
discovery	B-Algorithm
system	I-Algorithm
is	O
an	O
artificial	B-Application
intelligence	I-Application
system	O
that	O
attempts	O
to	O
discover	O
new	O
scientific	O
concepts	O
or	O
laws	O
.	O
</s>
<s>
The	O
aim	O
of	O
discovery	B-Algorithm
systems	I-Algorithm
is	O
to	O
automate	O
scientific	O
data	O
analysis	O
and	O
the	O
scientific	O
discovery	O
process	O
.	O
</s>
<s>
Ideally	O
,	O
an	O
artificial	B-Application
intelligence	I-Application
system	O
should	O
be	O
able	O
to	O
search	O
systematically	O
through	O
the	O
space	O
of	O
all	O
possible	O
hypotheses	O
and	O
yield	O
the	O
hypothesis	O
-	O
or	O
set	O
of	O
equally	O
likely	O
hypotheses	O
-	O
that	O
best	O
describes	O
the	O
complex	O
patterns	O
in	O
data	O
.	O
</s>
<s>
During	O
the	O
era	O
known	O
as	O
the	O
second	O
AI	B-Application
summer	O
(	O
approximately	O
1978-1987	O
)	O
,	O
various	O
systems	O
akin	O
to	O
the	O
era	O
's	O
dominant	O
expert	B-Application
systems	I-Application
were	O
developed	O
to	O
tackle	O
the	O
problem	O
of	O
extracting	O
scientific	O
hypotheses	O
from	O
data	O
,	O
with	O
or	O
without	O
interacting	O
with	O
a	O
human	O
scientist	O
.	O
</s>
<s>
These	O
systems	O
included	O
Autoclass	O
,	O
Automated	B-General_Concept
Mathematician	I-General_Concept
,	O
Eurisko	B-Algorithm
,	O
which	O
aimed	O
at	O
general-purpose	O
hypothesis	O
discovery	O
,	O
and	O
more	O
specific	O
systems	O
such	O
as	O
Dalton	B-Application
,	O
which	O
uncovers	O
molecular	O
properties	O
from	O
data	O
.	O
</s>
<s>
The	O
dream	O
of	O
building	O
systems	O
that	O
discover	O
scientific	O
hypotheses	O
was	O
pushed	O
to	O
the	O
background	O
with	O
the	O
second	O
AI	B-Application
winter	O
and	O
the	O
subsequent	O
resurgence	O
of	O
subsymbolic	O
methods	O
such	O
as	O
neural	B-Architecture
networks	I-Architecture
.	O
</s>
<s>
Subsymbolic	O
methods	O
emphasize	O
prediction	O
over	O
explanation	O
,	O
and	O
yield	O
models	O
which	O
works	O
well	O
but	O
are	O
difficult	O
or	O
impossible	O
to	O
explain	O
which	O
has	O
earned	O
them	O
the	O
name	O
black	B-Device
box	I-Device
AI	B-Application
.	O
</s>
<s>
Automated	B-General_Concept
Mathematician	I-General_Concept
was	O
one	O
of	O
the	O
earliest	O
successful	O
discovery	B-Algorithm
systems	I-Algorithm
.	O
</s>
<s>
After	O
a	O
couple	O
of	O
decades	O
with	O
little	O
interest	O
in	O
discovery	B-Algorithm
systems	I-Algorithm
,	O
the	O
interest	O
in	O
using	O
AI	B-Application
to	O
uncover	O
natural	O
laws	O
and	O
scientific	O
explanations	O
was	O
renewed	O
by	O
the	O
work	O
of	O
Michael	O
Schmidt	O
,	O
then	O
a	O
PhD	O
student	O
in	O
Computational	O
Biology	O
at	O
Cornell	O
University	O
.	O
</s>
<s>
Schmidt	O
and	O
his	O
advisor	O
,	O
Hod	O
Lipson	O
,	O
invented	O
Eureqa	B-Algorithm
,	O
which	O
they	O
described	O
as	O
a	O
symbolic	B-Algorithm
regression	I-Algorithm
approach	O
to	O
"	O
distilling	O
free-form	O
natural	O
laws	O
from	O
experimental	O
data	O
"	O
.	O
</s>
<s>
This	O
work	O
effectively	O
demonstrated	O
that	O
symbolic	B-Algorithm
regression	I-Algorithm
was	O
a	O
promising	O
way	O
forward	O
for	O
AI-driven	O
scientific	O
discovery	O
.	O
</s>
<s>
Since	O
2009	O
,	O
symbolic	B-Algorithm
regression	I-Algorithm
has	O
matured	O
further	O
,	O
and	O
today	O
,	O
various	O
commercial	O
and	O
open	O
source	O
systems	O
are	O
actively	O
used	O
in	O
scientific	O
research	O
.	O
</s>
<s>
Notable	O
examples	O
include	O
Eureqa	B-Algorithm
,	O
now	O
a	O
part	O
of	O
DataRobot	O
AI	B-Application
Cloud	O
Platform	O
,	O
AI	B-Application
Feynman	O
,	O
and	O
QLattice	B-Algorithm
.	O
</s>
